xref: /linux/net/ipv6/ip6_fib.c (revision 92ce4c3ea7c44e61ca2b6ef3e5682bfcea851d87)
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 #define RT6_DEBUG 2
42 
43 #if RT6_DEBUG >= 3
44 #define RT6_TRACE(x...) pr_debug(x)
45 #else
46 #define RT6_TRACE(x...) do { ; } while (0)
47 #endif
48 
49 static struct kmem_cache *fib6_node_kmem __read_mostly;
50 
51 struct fib6_cleaner {
52 	struct fib6_walker w;
53 	struct net *net;
54 	int (*func)(struct rt6_info *, void *arg);
55 	int sernum;
56 	void *arg;
57 };
58 
59 #ifdef CONFIG_IPV6_SUBTREES
60 #define FWS_INIT FWS_S
61 #else
62 #define FWS_INIT FWS_L
63 #endif
64 
65 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
66 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
67 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
68 static int fib6_walk(struct net *net, struct fib6_walker *w);
69 static int fib6_walk_continue(struct fib6_walker *w);
70 
71 /*
72  *	A routing update causes an increase of the serial number on the
73  *	affected subtree. This allows for cached routes to be asynchronously
74  *	tested when modifications are made to the destination cache as a
75  *	result of redirects, path MTU changes, etc.
76  */
77 
78 static void fib6_gc_timer_cb(unsigned long arg);
79 
80 #define FOR_WALKERS(net, w) \
81 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
82 
83 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
84 {
85 	write_lock_bh(&net->ipv6.fib6_walker_lock);
86 	list_add(&w->lh, &net->ipv6.fib6_walkers);
87 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
88 }
89 
90 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
91 {
92 	write_lock_bh(&net->ipv6.fib6_walker_lock);
93 	list_del(&w->lh);
94 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
95 }
96 
97 static int fib6_new_sernum(struct net *net)
98 {
99 	int new, old;
100 
101 	do {
102 		old = atomic_read(&net->ipv6.fib6_sernum);
103 		new = old < INT_MAX ? old + 1 : 1;
104 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
105 				old, new) != old);
106 	return new;
107 }
108 
109 enum {
110 	FIB6_NO_SERNUM_CHANGE = 0,
111 };
112 
113 /*
114  *	Auxiliary address test functions for the radix tree.
115  *
116  *	These assume a 32bit processor (although it will work on
117  *	64bit processors)
118  */
119 
120 /*
121  *	test bit
122  */
123 #if defined(__LITTLE_ENDIAN)
124 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
125 #else
126 # define BITOP_BE32_SWIZZLE	0
127 #endif
128 
129 static __be32 addr_bit_set(const void *token, int fn_bit)
130 {
131 	const __be32 *addr = token;
132 	/*
133 	 * Here,
134 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
135 	 * is optimized version of
136 	 *	htonl(1 << ((~fn_bit)&0x1F))
137 	 * See include/asm-generic/bitops/le.h.
138 	 */
139 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
140 	       addr[fn_bit >> 5];
141 }
142 
143 static struct fib6_node *node_alloc(void)
144 {
145 	struct fib6_node *fn;
146 
147 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
148 
149 	return fn;
150 }
151 
152 static void node_free_immediate(struct fib6_node *fn)
153 {
154 	kmem_cache_free(fib6_node_kmem, fn);
155 }
156 
157 static void node_free_rcu(struct rcu_head *head)
158 {
159 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
160 
161 	kmem_cache_free(fib6_node_kmem, fn);
162 }
163 
164 static void node_free(struct fib6_node *fn)
165 {
166 	call_rcu(&fn->rcu, node_free_rcu);
167 }
168 
169 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
170 {
171 	int cpu;
172 
173 	if (!non_pcpu_rt->rt6i_pcpu)
174 		return;
175 
176 	for_each_possible_cpu(cpu) {
177 		struct rt6_info **ppcpu_rt;
178 		struct rt6_info *pcpu_rt;
179 
180 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
181 		pcpu_rt = *ppcpu_rt;
182 		if (pcpu_rt) {
183 			dst_dev_put(&pcpu_rt->dst);
184 			dst_release(&pcpu_rt->dst);
185 			*ppcpu_rt = NULL;
186 		}
187 	}
188 
189 	free_percpu(non_pcpu_rt->rt6i_pcpu);
190 	non_pcpu_rt->rt6i_pcpu = NULL;
191 }
192 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
193 
194 static void fib6_link_table(struct net *net, struct fib6_table *tb)
195 {
196 	unsigned int h;
197 
198 	/*
199 	 * Initialize table lock at a single place to give lockdep a key,
200 	 * tables aren't visible prior to being linked to the list.
201 	 */
202 	rwlock_init(&tb->tb6_lock);
203 
204 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
205 
206 	/*
207 	 * No protection necessary, this is the only list mutatation
208 	 * operation, tables never disappear once they exist.
209 	 */
210 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
211 }
212 
213 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
214 
215 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
216 {
217 	struct fib6_table *table;
218 
219 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
220 	if (table) {
221 		table->tb6_id = id;
222 		table->tb6_root.leaf = net->ipv6.ip6_null_entry;
223 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
224 		inet_peer_base_init(&table->tb6_peers);
225 	}
226 
227 	return table;
228 }
229 
230 struct fib6_table *fib6_new_table(struct net *net, u32 id)
231 {
232 	struct fib6_table *tb;
233 
234 	if (id == 0)
235 		id = RT6_TABLE_MAIN;
236 	tb = fib6_get_table(net, id);
237 	if (tb)
238 		return tb;
239 
240 	tb = fib6_alloc_table(net, id);
241 	if (tb)
242 		fib6_link_table(net, tb);
243 
244 	return tb;
245 }
246 EXPORT_SYMBOL_GPL(fib6_new_table);
247 
248 struct fib6_table *fib6_get_table(struct net *net, u32 id)
249 {
250 	struct fib6_table *tb;
251 	struct hlist_head *head;
252 	unsigned int h;
253 
254 	if (id == 0)
255 		id = RT6_TABLE_MAIN;
256 	h = id & (FIB6_TABLE_HASHSZ - 1);
257 	rcu_read_lock();
258 	head = &net->ipv6.fib_table_hash[h];
259 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
260 		if (tb->tb6_id == id) {
261 			rcu_read_unlock();
262 			return tb;
263 		}
264 	}
265 	rcu_read_unlock();
266 
267 	return NULL;
268 }
269 EXPORT_SYMBOL_GPL(fib6_get_table);
270 
271 static void __net_init fib6_tables_init(struct net *net)
272 {
273 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
274 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
275 }
276 #else
277 
278 struct fib6_table *fib6_new_table(struct net *net, u32 id)
279 {
280 	return fib6_get_table(net, id);
281 }
282 
283 struct fib6_table *fib6_get_table(struct net *net, u32 id)
284 {
285 	  return net->ipv6.fib6_main_tbl;
286 }
287 
288 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
289 				   int flags, pol_lookup_t lookup)
290 {
291 	struct rt6_info *rt;
292 
293 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
294 	if (rt->dst.error == -EAGAIN) {
295 		ip6_rt_put(rt);
296 		rt = net->ipv6.ip6_null_entry;
297 		dst_hold(&rt->dst);
298 	}
299 
300 	return &rt->dst;
301 }
302 
303 static void __net_init fib6_tables_init(struct net *net)
304 {
305 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
306 }
307 
308 #endif
309 
310 unsigned int fib6_tables_seq_read(struct net *net)
311 {
312 	unsigned int h, fib_seq = 0;
313 
314 	rcu_read_lock();
315 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
316 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
317 		struct fib6_table *tb;
318 
319 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
320 			read_lock_bh(&tb->tb6_lock);
321 			fib_seq += tb->fib_seq;
322 			read_unlock_bh(&tb->tb6_lock);
323 		}
324 	}
325 	rcu_read_unlock();
326 
327 	return fib_seq;
328 }
329 
330 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
331 				    enum fib_event_type event_type,
332 				    struct rt6_info *rt)
333 {
334 	struct fib6_entry_notifier_info info = {
335 		.rt = rt,
336 	};
337 
338 	return call_fib6_notifier(nb, net, event_type, &info.info);
339 }
340 
341 static int call_fib6_entry_notifiers(struct net *net,
342 				     enum fib_event_type event_type,
343 				     struct rt6_info *rt)
344 {
345 	struct fib6_entry_notifier_info info = {
346 		.rt = rt,
347 	};
348 
349 	rt->rt6i_table->fib_seq++;
350 	return call_fib6_notifiers(net, event_type, &info.info);
351 }
352 
353 struct fib6_dump_arg {
354 	struct net *net;
355 	struct notifier_block *nb;
356 };
357 
358 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
359 {
360 	if (rt == arg->net->ipv6.ip6_null_entry)
361 		return;
362 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
363 }
364 
365 static int fib6_node_dump(struct fib6_walker *w)
366 {
367 	struct rt6_info *rt;
368 
369 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next)
370 		fib6_rt_dump(rt, w->args);
371 	w->leaf = NULL;
372 	return 0;
373 }
374 
375 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
376 			    struct fib6_walker *w)
377 {
378 	w->root = &tb->tb6_root;
379 	read_lock_bh(&tb->tb6_lock);
380 	fib6_walk(net, w);
381 	read_unlock_bh(&tb->tb6_lock);
382 }
383 
384 /* Called with rcu_read_lock() */
385 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
386 {
387 	struct fib6_dump_arg arg;
388 	struct fib6_walker *w;
389 	unsigned int h;
390 
391 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
392 	if (!w)
393 		return -ENOMEM;
394 
395 	w->func = fib6_node_dump;
396 	arg.net = net;
397 	arg.nb = nb;
398 	w->args = &arg;
399 
400 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
401 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
402 		struct fib6_table *tb;
403 
404 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
405 			fib6_table_dump(net, tb, w);
406 	}
407 
408 	kfree(w);
409 
410 	return 0;
411 }
412 
413 static int fib6_dump_node(struct fib6_walker *w)
414 {
415 	int res;
416 	struct rt6_info *rt;
417 
418 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
419 		res = rt6_dump_route(rt, w->args);
420 		if (res < 0) {
421 			/* Frame is full, suspend walking */
422 			w->leaf = rt;
423 			return 1;
424 		}
425 
426 		/* Multipath routes are dumped in one route with the
427 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
428 		 * last sibling of this route (no need to dump the
429 		 * sibling routes again)
430 		 */
431 		if (rt->rt6i_nsiblings)
432 			rt = list_last_entry(&rt->rt6i_siblings,
433 					     struct rt6_info,
434 					     rt6i_siblings);
435 	}
436 	w->leaf = NULL;
437 	return 0;
438 }
439 
440 static void fib6_dump_end(struct netlink_callback *cb)
441 {
442 	struct net *net = sock_net(cb->skb->sk);
443 	struct fib6_walker *w = (void *)cb->args[2];
444 
445 	if (w) {
446 		if (cb->args[4]) {
447 			cb->args[4] = 0;
448 			fib6_walker_unlink(net, w);
449 		}
450 		cb->args[2] = 0;
451 		kfree(w);
452 	}
453 	cb->done = (void *)cb->args[3];
454 	cb->args[1] = 3;
455 }
456 
457 static int fib6_dump_done(struct netlink_callback *cb)
458 {
459 	fib6_dump_end(cb);
460 	return cb->done ? cb->done(cb) : 0;
461 }
462 
463 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
464 			   struct netlink_callback *cb)
465 {
466 	struct net *net = sock_net(skb->sk);
467 	struct fib6_walker *w;
468 	int res;
469 
470 	w = (void *)cb->args[2];
471 	w->root = &table->tb6_root;
472 
473 	if (cb->args[4] == 0) {
474 		w->count = 0;
475 		w->skip = 0;
476 
477 		read_lock_bh(&table->tb6_lock);
478 		res = fib6_walk(net, w);
479 		read_unlock_bh(&table->tb6_lock);
480 		if (res > 0) {
481 			cb->args[4] = 1;
482 			cb->args[5] = w->root->fn_sernum;
483 		}
484 	} else {
485 		if (cb->args[5] != w->root->fn_sernum) {
486 			/* Begin at the root if the tree changed */
487 			cb->args[5] = w->root->fn_sernum;
488 			w->state = FWS_INIT;
489 			w->node = w->root;
490 			w->skip = w->count;
491 		} else
492 			w->skip = 0;
493 
494 		read_lock_bh(&table->tb6_lock);
495 		res = fib6_walk_continue(w);
496 		read_unlock_bh(&table->tb6_lock);
497 		if (res <= 0) {
498 			fib6_walker_unlink(net, w);
499 			cb->args[4] = 0;
500 		}
501 	}
502 
503 	return res;
504 }
505 
506 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
507 {
508 	struct net *net = sock_net(skb->sk);
509 	unsigned int h, s_h;
510 	unsigned int e = 0, s_e;
511 	struct rt6_rtnl_dump_arg arg;
512 	struct fib6_walker *w;
513 	struct fib6_table *tb;
514 	struct hlist_head *head;
515 	int res = 0;
516 
517 	s_h = cb->args[0];
518 	s_e = cb->args[1];
519 
520 	w = (void *)cb->args[2];
521 	if (!w) {
522 		/* New dump:
523 		 *
524 		 * 1. hook callback destructor.
525 		 */
526 		cb->args[3] = (long)cb->done;
527 		cb->done = fib6_dump_done;
528 
529 		/*
530 		 * 2. allocate and initialize walker.
531 		 */
532 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
533 		if (!w)
534 			return -ENOMEM;
535 		w->func = fib6_dump_node;
536 		cb->args[2] = (long)w;
537 	}
538 
539 	arg.skb = skb;
540 	arg.cb = cb;
541 	arg.net = net;
542 	w->args = &arg;
543 
544 	rcu_read_lock();
545 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
546 		e = 0;
547 		head = &net->ipv6.fib_table_hash[h];
548 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
549 			if (e < s_e)
550 				goto next;
551 			res = fib6_dump_table(tb, skb, cb);
552 			if (res != 0)
553 				goto out;
554 next:
555 			e++;
556 		}
557 	}
558 out:
559 	rcu_read_unlock();
560 	cb->args[1] = e;
561 	cb->args[0] = h;
562 
563 	res = res < 0 ? res : skb->len;
564 	if (res <= 0)
565 		fib6_dump_end(cb);
566 	return res;
567 }
568 
569 /*
570  *	Routing Table
571  *
572  *	return the appropriate node for a routing tree "add" operation
573  *	by either creating and inserting or by returning an existing
574  *	node.
575  */
576 
577 static struct fib6_node *fib6_add_1(struct fib6_node *root,
578 				     struct in6_addr *addr, int plen,
579 				     int offset, int allow_create,
580 				     int replace_required, int sernum,
581 				     struct netlink_ext_ack *extack)
582 {
583 	struct fib6_node *fn, *in, *ln;
584 	struct fib6_node *pn = NULL;
585 	struct rt6key *key;
586 	int	bit;
587 	__be32	dir = 0;
588 
589 	RT6_TRACE("fib6_add_1\n");
590 
591 	/* insert node in tree */
592 
593 	fn = root;
594 
595 	do {
596 		key = (struct rt6key *)((u8 *)fn->leaf + offset);
597 
598 		/*
599 		 *	Prefix match
600 		 */
601 		if (plen < fn->fn_bit ||
602 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
603 			if (!allow_create) {
604 				if (replace_required) {
605 					NL_SET_ERR_MSG(extack,
606 						       "Can not replace route - no match found");
607 					pr_warn("Can't replace route, no match found\n");
608 					return ERR_PTR(-ENOENT);
609 				}
610 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
611 			}
612 			goto insert_above;
613 		}
614 
615 		/*
616 		 *	Exact match ?
617 		 */
618 
619 		if (plen == fn->fn_bit) {
620 			/* clean up an intermediate node */
621 			if (!(fn->fn_flags & RTN_RTINFO)) {
622 				rt6_release(fn->leaf);
623 				fn->leaf = NULL;
624 			}
625 
626 			fn->fn_sernum = sernum;
627 
628 			return fn;
629 		}
630 
631 		/*
632 		 *	We have more bits to go
633 		 */
634 
635 		/* Try to walk down on tree. */
636 		fn->fn_sernum = sernum;
637 		dir = addr_bit_set(addr, fn->fn_bit);
638 		pn = fn;
639 		fn = dir ? fn->right : fn->left;
640 	} while (fn);
641 
642 	if (!allow_create) {
643 		/* We should not create new node because
644 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
645 		 * I assume it is safe to require NLM_F_CREATE when
646 		 * REPLACE flag is used! Later we may want to remove the
647 		 * check for replace_required, because according
648 		 * to netlink specification, NLM_F_CREATE
649 		 * MUST be specified if new route is created.
650 		 * That would keep IPv6 consistent with IPv4
651 		 */
652 		if (replace_required) {
653 			NL_SET_ERR_MSG(extack,
654 				       "Can not replace route - no match found");
655 			pr_warn("Can't replace route, no match found\n");
656 			return ERR_PTR(-ENOENT);
657 		}
658 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
659 	}
660 	/*
661 	 *	We walked to the bottom of tree.
662 	 *	Create new leaf node without children.
663 	 */
664 
665 	ln = node_alloc();
666 
667 	if (!ln)
668 		return ERR_PTR(-ENOMEM);
669 	ln->fn_bit = plen;
670 
671 	ln->parent = pn;
672 	ln->fn_sernum = sernum;
673 
674 	if (dir)
675 		pn->right = ln;
676 	else
677 		pn->left  = ln;
678 
679 	return ln;
680 
681 
682 insert_above:
683 	/*
684 	 * split since we don't have a common prefix anymore or
685 	 * we have a less significant route.
686 	 * we've to insert an intermediate node on the list
687 	 * this new node will point to the one we need to create
688 	 * and the current
689 	 */
690 
691 	pn = fn->parent;
692 
693 	/* find 1st bit in difference between the 2 addrs.
694 
695 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
696 	   but if it is >= plen, the value is ignored in any case.
697 	 */
698 
699 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
700 
701 	/*
702 	 *		(intermediate)[in]
703 	 *	          /	   \
704 	 *	(new leaf node)[ln] (old node)[fn]
705 	 */
706 	if (plen > bit) {
707 		in = node_alloc();
708 		ln = node_alloc();
709 
710 		if (!in || !ln) {
711 			if (in)
712 				node_free_immediate(in);
713 			if (ln)
714 				node_free_immediate(ln);
715 			return ERR_PTR(-ENOMEM);
716 		}
717 
718 		/*
719 		 * new intermediate node.
720 		 * RTN_RTINFO will
721 		 * be off since that an address that chooses one of
722 		 * the branches would not match less specific routes
723 		 * in the other branch
724 		 */
725 
726 		in->fn_bit = bit;
727 
728 		in->parent = pn;
729 		in->leaf = fn->leaf;
730 		atomic_inc(&in->leaf->rt6i_ref);
731 
732 		in->fn_sernum = sernum;
733 
734 		/* update parent pointer */
735 		if (dir)
736 			pn->right = in;
737 		else
738 			pn->left  = in;
739 
740 		ln->fn_bit = plen;
741 
742 		ln->parent = in;
743 		fn->parent = in;
744 
745 		ln->fn_sernum = sernum;
746 
747 		if (addr_bit_set(addr, bit)) {
748 			in->right = ln;
749 			in->left  = fn;
750 		} else {
751 			in->left  = ln;
752 			in->right = fn;
753 		}
754 	} else { /* plen <= bit */
755 
756 		/*
757 		 *		(new leaf node)[ln]
758 		 *	          /	   \
759 		 *	     (old node)[fn] NULL
760 		 */
761 
762 		ln = node_alloc();
763 
764 		if (!ln)
765 			return ERR_PTR(-ENOMEM);
766 
767 		ln->fn_bit = plen;
768 
769 		ln->parent = pn;
770 
771 		ln->fn_sernum = sernum;
772 
773 		if (dir)
774 			pn->right = ln;
775 		else
776 			pn->left  = ln;
777 
778 		if (addr_bit_set(&key->addr, plen))
779 			ln->right = fn;
780 		else
781 			ln->left  = fn;
782 
783 		fn->parent = ln;
784 	}
785 	return ln;
786 }
787 
788 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
789 {
790 	return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
791 	       RTF_GATEWAY;
792 }
793 
794 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
795 {
796 	int i;
797 
798 	for (i = 0; i < RTAX_MAX; i++) {
799 		if (test_bit(i, mxc->mx_valid))
800 			mp[i] = mxc->mx[i];
801 	}
802 }
803 
804 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
805 {
806 	if (!mxc->mx)
807 		return 0;
808 
809 	if (dst->flags & DST_HOST) {
810 		u32 *mp = dst_metrics_write_ptr(dst);
811 
812 		if (unlikely(!mp))
813 			return -ENOMEM;
814 
815 		fib6_copy_metrics(mp, mxc);
816 	} else {
817 		dst_init_metrics(dst, mxc->mx, false);
818 
819 		/* We've stolen mx now. */
820 		mxc->mx = NULL;
821 	}
822 
823 	return 0;
824 }
825 
826 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
827 			  struct net *net)
828 {
829 	if (atomic_read(&rt->rt6i_ref) != 1) {
830 		/* This route is used as dummy address holder in some split
831 		 * nodes. It is not leaked, but it still holds other resources,
832 		 * which must be released in time. So, scan ascendant nodes
833 		 * and replace dummy references to this route with references
834 		 * to still alive ones.
835 		 */
836 		while (fn) {
837 			if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
838 				fn->leaf = fib6_find_prefix(net, fn);
839 				atomic_inc(&fn->leaf->rt6i_ref);
840 				rt6_release(rt);
841 			}
842 			fn = fn->parent;
843 		}
844 	}
845 }
846 
847 /*
848  *	Insert routing information in a node.
849  */
850 
851 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
852 			    struct nl_info *info, struct mx6_config *mxc)
853 {
854 	struct rt6_info *iter = NULL;
855 	struct rt6_info **ins;
856 	struct rt6_info **fallback_ins = NULL;
857 	int replace = (info->nlh &&
858 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
859 	int add = (!info->nlh ||
860 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
861 	int found = 0;
862 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
863 	u16 nlflags = NLM_F_EXCL;
864 	int err;
865 
866 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
867 		nlflags |= NLM_F_APPEND;
868 
869 	ins = &fn->leaf;
870 
871 	for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
872 		/*
873 		 *	Search for duplicates
874 		 */
875 
876 		if (iter->rt6i_metric == rt->rt6i_metric) {
877 			/*
878 			 *	Same priority level
879 			 */
880 			if (info->nlh &&
881 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
882 				return -EEXIST;
883 
884 			nlflags &= ~NLM_F_EXCL;
885 			if (replace) {
886 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
887 					found++;
888 					break;
889 				}
890 				if (rt_can_ecmp)
891 					fallback_ins = fallback_ins ?: ins;
892 				goto next_iter;
893 			}
894 
895 			if (rt6_duplicate_nexthop(iter, rt)) {
896 				if (rt->rt6i_nsiblings)
897 					rt->rt6i_nsiblings = 0;
898 				if (!(iter->rt6i_flags & RTF_EXPIRES))
899 					return -EEXIST;
900 				if (!(rt->rt6i_flags & RTF_EXPIRES))
901 					rt6_clean_expires(iter);
902 				else
903 					rt6_set_expires(iter, rt->dst.expires);
904 				iter->rt6i_pmtu = rt->rt6i_pmtu;
905 				return -EEXIST;
906 			}
907 			/* If we have the same destination and the same metric,
908 			 * but not the same gateway, then the route we try to
909 			 * add is sibling to this route, increment our counter
910 			 * of siblings, and later we will add our route to the
911 			 * list.
912 			 * Only static routes (which don't have flag
913 			 * RTF_EXPIRES) are used for ECMPv6.
914 			 *
915 			 * To avoid long list, we only had siblings if the
916 			 * route have a gateway.
917 			 */
918 			if (rt_can_ecmp &&
919 			    rt6_qualify_for_ecmp(iter))
920 				rt->rt6i_nsiblings++;
921 		}
922 
923 		if (iter->rt6i_metric > rt->rt6i_metric)
924 			break;
925 
926 next_iter:
927 		ins = &iter->dst.rt6_next;
928 	}
929 
930 	if (fallback_ins && !found) {
931 		/* No ECMP-able route found, replace first non-ECMP one */
932 		ins = fallback_ins;
933 		iter = *ins;
934 		found++;
935 	}
936 
937 	/* Reset round-robin state, if necessary */
938 	if (ins == &fn->leaf)
939 		fn->rr_ptr = NULL;
940 
941 	/* Link this route to others same route. */
942 	if (rt->rt6i_nsiblings) {
943 		unsigned int rt6i_nsiblings;
944 		struct rt6_info *sibling, *temp_sibling;
945 
946 		/* Find the first route that have the same metric */
947 		sibling = fn->leaf;
948 		while (sibling) {
949 			if (sibling->rt6i_metric == rt->rt6i_metric &&
950 			    rt6_qualify_for_ecmp(sibling)) {
951 				list_add_tail(&rt->rt6i_siblings,
952 					      &sibling->rt6i_siblings);
953 				break;
954 			}
955 			sibling = sibling->dst.rt6_next;
956 		}
957 		/* For each sibling in the list, increment the counter of
958 		 * siblings. BUG() if counters does not match, list of siblings
959 		 * is broken!
960 		 */
961 		rt6i_nsiblings = 0;
962 		list_for_each_entry_safe(sibling, temp_sibling,
963 					 &rt->rt6i_siblings, rt6i_siblings) {
964 			sibling->rt6i_nsiblings++;
965 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
966 			rt6i_nsiblings++;
967 		}
968 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
969 	}
970 
971 	/*
972 	 *	insert node
973 	 */
974 	if (!replace) {
975 		if (!add)
976 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
977 
978 add:
979 		nlflags |= NLM_F_CREATE;
980 		err = fib6_commit_metrics(&rt->dst, mxc);
981 		if (err)
982 			return err;
983 
984 		rt->dst.rt6_next = iter;
985 		*ins = rt;
986 		rcu_assign_pointer(rt->rt6i_node, fn);
987 		atomic_inc(&rt->rt6i_ref);
988 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
989 					  rt);
990 		if (!info->skip_notify)
991 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
992 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
993 
994 		if (!(fn->fn_flags & RTN_RTINFO)) {
995 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
996 			fn->fn_flags |= RTN_RTINFO;
997 		}
998 
999 	} else {
1000 		int nsiblings;
1001 
1002 		if (!found) {
1003 			if (add)
1004 				goto add;
1005 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1006 			return -ENOENT;
1007 		}
1008 
1009 		err = fib6_commit_metrics(&rt->dst, mxc);
1010 		if (err)
1011 			return err;
1012 
1013 		*ins = rt;
1014 		rcu_assign_pointer(rt->rt6i_node, fn);
1015 		rt->dst.rt6_next = iter->dst.rt6_next;
1016 		atomic_inc(&rt->rt6i_ref);
1017 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1018 					  rt);
1019 		if (!info->skip_notify)
1020 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1021 		if (!(fn->fn_flags & RTN_RTINFO)) {
1022 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1023 			fn->fn_flags |= RTN_RTINFO;
1024 		}
1025 		nsiblings = iter->rt6i_nsiblings;
1026 		iter->rt6i_node = NULL;
1027 		fib6_purge_rt(iter, fn, info->nl_net);
1028 		if (fn->rr_ptr == iter)
1029 			fn->rr_ptr = NULL;
1030 		rt6_release(iter);
1031 
1032 		if (nsiblings) {
1033 			/* Replacing an ECMP route, remove all siblings */
1034 			ins = &rt->dst.rt6_next;
1035 			iter = *ins;
1036 			while (iter) {
1037 				if (iter->rt6i_metric > rt->rt6i_metric)
1038 					break;
1039 				if (rt6_qualify_for_ecmp(iter)) {
1040 					*ins = iter->dst.rt6_next;
1041 					iter->rt6i_node = NULL;
1042 					fib6_purge_rt(iter, fn, info->nl_net);
1043 					if (fn->rr_ptr == iter)
1044 						fn->rr_ptr = NULL;
1045 					rt6_release(iter);
1046 					nsiblings--;
1047 				} else {
1048 					ins = &iter->dst.rt6_next;
1049 				}
1050 				iter = *ins;
1051 			}
1052 			WARN_ON(nsiblings != 0);
1053 		}
1054 	}
1055 
1056 	return 0;
1057 }
1058 
1059 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1060 {
1061 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1062 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1063 		mod_timer(&net->ipv6.ip6_fib_timer,
1064 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1065 }
1066 
1067 void fib6_force_start_gc(struct net *net)
1068 {
1069 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1070 		mod_timer(&net->ipv6.ip6_fib_timer,
1071 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1072 }
1073 
1074 /*
1075  *	Add routing information to the routing tree.
1076  *	<destination addr>/<source addr>
1077  *	with source addr info in sub-trees
1078  */
1079 
1080 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1081 	     struct nl_info *info, struct mx6_config *mxc,
1082 	     struct netlink_ext_ack *extack)
1083 {
1084 	struct fib6_node *fn, *pn = NULL;
1085 	int err = -ENOMEM;
1086 	int allow_create = 1;
1087 	int replace_required = 0;
1088 	int sernum = fib6_new_sernum(info->nl_net);
1089 
1090 	if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1091 		return -EINVAL;
1092 
1093 	if (info->nlh) {
1094 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1095 			allow_create = 0;
1096 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1097 			replace_required = 1;
1098 	}
1099 	if (!allow_create && !replace_required)
1100 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1101 
1102 	fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1103 			offsetof(struct rt6_info, rt6i_dst), allow_create,
1104 			replace_required, sernum, extack);
1105 	if (IS_ERR(fn)) {
1106 		err = PTR_ERR(fn);
1107 		fn = NULL;
1108 		goto out;
1109 	}
1110 
1111 	pn = fn;
1112 
1113 #ifdef CONFIG_IPV6_SUBTREES
1114 	if (rt->rt6i_src.plen) {
1115 		struct fib6_node *sn;
1116 
1117 		if (!fn->subtree) {
1118 			struct fib6_node *sfn;
1119 
1120 			/*
1121 			 * Create subtree.
1122 			 *
1123 			 *		fn[main tree]
1124 			 *		|
1125 			 *		sfn[subtree root]
1126 			 *		   \
1127 			 *		    sn[new leaf node]
1128 			 */
1129 
1130 			/* Create subtree root node */
1131 			sfn = node_alloc();
1132 			if (!sfn)
1133 				goto failure;
1134 
1135 			sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1136 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1137 			sfn->fn_flags = RTN_ROOT;
1138 			sfn->fn_sernum = sernum;
1139 
1140 			/* Now add the first leaf node to new subtree */
1141 
1142 			sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1143 					rt->rt6i_src.plen,
1144 					offsetof(struct rt6_info, rt6i_src),
1145 					allow_create, replace_required, sernum,
1146 					extack);
1147 
1148 			if (IS_ERR(sn)) {
1149 				/* If it is failed, discard just allocated
1150 				   root, and then (in failure) stale node
1151 				   in main tree.
1152 				 */
1153 				node_free_immediate(sfn);
1154 				err = PTR_ERR(sn);
1155 				goto failure;
1156 			}
1157 
1158 			/* Now link new subtree to main tree */
1159 			sfn->parent = fn;
1160 			fn->subtree = sfn;
1161 		} else {
1162 			sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1163 					rt->rt6i_src.plen,
1164 					offsetof(struct rt6_info, rt6i_src),
1165 					allow_create, replace_required, sernum,
1166 					extack);
1167 
1168 			if (IS_ERR(sn)) {
1169 				err = PTR_ERR(sn);
1170 				goto failure;
1171 			}
1172 		}
1173 
1174 		if (!fn->leaf) {
1175 			fn->leaf = rt;
1176 			atomic_inc(&rt->rt6i_ref);
1177 		}
1178 		fn = sn;
1179 	}
1180 #endif
1181 
1182 	err = fib6_add_rt2node(fn, rt, info, mxc);
1183 	if (!err) {
1184 		fib6_start_gc(info->nl_net, rt);
1185 		if (!(rt->rt6i_flags & RTF_CACHE))
1186 			fib6_prune_clones(info->nl_net, pn);
1187 	}
1188 
1189 out:
1190 	if (err) {
1191 #ifdef CONFIG_IPV6_SUBTREES
1192 		/*
1193 		 * If fib6_add_1 has cleared the old leaf pointer in the
1194 		 * super-tree leaf node we have to find a new one for it.
1195 		 */
1196 		if (pn != fn && pn->leaf == rt) {
1197 			pn->leaf = NULL;
1198 			atomic_dec(&rt->rt6i_ref);
1199 		}
1200 		if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1201 			pn->leaf = fib6_find_prefix(info->nl_net, pn);
1202 #if RT6_DEBUG >= 2
1203 			if (!pn->leaf) {
1204 				WARN_ON(pn->leaf == NULL);
1205 				pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1206 			}
1207 #endif
1208 			atomic_inc(&pn->leaf->rt6i_ref);
1209 		}
1210 #endif
1211 		goto failure;
1212 	}
1213 	return err;
1214 
1215 failure:
1216 	/* fn->leaf could be NULL if fn is an intermediate node and we
1217 	 * failed to add the new route to it in both subtree creation
1218 	 * failure and fib6_add_rt2node() failure case.
1219 	 * In both cases, fib6_repair_tree() should be called to fix
1220 	 * fn->leaf.
1221 	 */
1222 	if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1223 		fib6_repair_tree(info->nl_net, fn);
1224 	/* Always release dst as dst->__refcnt is guaranteed
1225 	 * to be taken before entering this function
1226 	 */
1227 	dst_release_immediate(&rt->dst);
1228 	return err;
1229 }
1230 
1231 /*
1232  *	Routing tree lookup
1233  *
1234  */
1235 
1236 struct lookup_args {
1237 	int			offset;		/* key offset on rt6_info	*/
1238 	const struct in6_addr	*addr;		/* search key			*/
1239 };
1240 
1241 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1242 				       struct lookup_args *args)
1243 {
1244 	struct fib6_node *fn;
1245 	__be32 dir;
1246 
1247 	if (unlikely(args->offset == 0))
1248 		return NULL;
1249 
1250 	/*
1251 	 *	Descend on a tree
1252 	 */
1253 
1254 	fn = root;
1255 
1256 	for (;;) {
1257 		struct fib6_node *next;
1258 
1259 		dir = addr_bit_set(args->addr, fn->fn_bit);
1260 
1261 		next = dir ? fn->right : fn->left;
1262 
1263 		if (next) {
1264 			fn = next;
1265 			continue;
1266 		}
1267 		break;
1268 	}
1269 
1270 	while (fn) {
1271 		if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1272 			struct rt6key *key;
1273 
1274 			key = (struct rt6key *) ((u8 *) fn->leaf +
1275 						 args->offset);
1276 
1277 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1278 #ifdef CONFIG_IPV6_SUBTREES
1279 				if (fn->subtree) {
1280 					struct fib6_node *sfn;
1281 					sfn = fib6_lookup_1(fn->subtree,
1282 							    args + 1);
1283 					if (!sfn)
1284 						goto backtrack;
1285 					fn = sfn;
1286 				}
1287 #endif
1288 				if (fn->fn_flags & RTN_RTINFO)
1289 					return fn;
1290 			}
1291 		}
1292 #ifdef CONFIG_IPV6_SUBTREES
1293 backtrack:
1294 #endif
1295 		if (fn->fn_flags & RTN_ROOT)
1296 			break;
1297 
1298 		fn = fn->parent;
1299 	}
1300 
1301 	return NULL;
1302 }
1303 
1304 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1305 			      const struct in6_addr *saddr)
1306 {
1307 	struct fib6_node *fn;
1308 	struct lookup_args args[] = {
1309 		{
1310 			.offset = offsetof(struct rt6_info, rt6i_dst),
1311 			.addr = daddr,
1312 		},
1313 #ifdef CONFIG_IPV6_SUBTREES
1314 		{
1315 			.offset = offsetof(struct rt6_info, rt6i_src),
1316 			.addr = saddr,
1317 		},
1318 #endif
1319 		{
1320 			.offset = 0,	/* sentinel */
1321 		}
1322 	};
1323 
1324 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1325 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1326 		fn = root;
1327 
1328 	return fn;
1329 }
1330 
1331 /*
1332  *	Get node with specified destination prefix (and source prefix,
1333  *	if subtrees are used)
1334  */
1335 
1336 
1337 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1338 				       const struct in6_addr *addr,
1339 				       int plen, int offset)
1340 {
1341 	struct fib6_node *fn;
1342 
1343 	for (fn = root; fn ; ) {
1344 		struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1345 
1346 		/*
1347 		 *	Prefix match
1348 		 */
1349 		if (plen < fn->fn_bit ||
1350 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1351 			return NULL;
1352 
1353 		if (plen == fn->fn_bit)
1354 			return fn;
1355 
1356 		/*
1357 		 *	We have more bits to go
1358 		 */
1359 		if (addr_bit_set(addr, fn->fn_bit))
1360 			fn = fn->right;
1361 		else
1362 			fn = fn->left;
1363 	}
1364 	return NULL;
1365 }
1366 
1367 struct fib6_node *fib6_locate(struct fib6_node *root,
1368 			      const struct in6_addr *daddr, int dst_len,
1369 			      const struct in6_addr *saddr, int src_len)
1370 {
1371 	struct fib6_node *fn;
1372 
1373 	fn = fib6_locate_1(root, daddr, dst_len,
1374 			   offsetof(struct rt6_info, rt6i_dst));
1375 
1376 #ifdef CONFIG_IPV6_SUBTREES
1377 	if (src_len) {
1378 		WARN_ON(saddr == NULL);
1379 		if (fn && fn->subtree)
1380 			fn = fib6_locate_1(fn->subtree, saddr, src_len,
1381 					   offsetof(struct rt6_info, rt6i_src));
1382 	}
1383 #endif
1384 
1385 	if (fn && fn->fn_flags & RTN_RTINFO)
1386 		return fn;
1387 
1388 	return NULL;
1389 }
1390 
1391 
1392 /*
1393  *	Deletion
1394  *
1395  */
1396 
1397 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1398 {
1399 	if (fn->fn_flags & RTN_ROOT)
1400 		return net->ipv6.ip6_null_entry;
1401 
1402 	while (fn) {
1403 		if (fn->left)
1404 			return fn->left->leaf;
1405 		if (fn->right)
1406 			return fn->right->leaf;
1407 
1408 		fn = FIB6_SUBTREE(fn);
1409 	}
1410 	return NULL;
1411 }
1412 
1413 /*
1414  *	Called to trim the tree of intermediate nodes when possible. "fn"
1415  *	is the node we want to try and remove.
1416  */
1417 
1418 static struct fib6_node *fib6_repair_tree(struct net *net,
1419 					   struct fib6_node *fn)
1420 {
1421 	int children;
1422 	int nstate;
1423 	struct fib6_node *child, *pn;
1424 	struct fib6_walker *w;
1425 	int iter = 0;
1426 
1427 	for (;;) {
1428 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1429 		iter++;
1430 
1431 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1432 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1433 		WARN_ON(fn->leaf);
1434 
1435 		children = 0;
1436 		child = NULL;
1437 		if (fn->right)
1438 			child = fn->right, children |= 1;
1439 		if (fn->left)
1440 			child = fn->left, children |= 2;
1441 
1442 		if (children == 3 || FIB6_SUBTREE(fn)
1443 #ifdef CONFIG_IPV6_SUBTREES
1444 		    /* Subtree root (i.e. fn) may have one child */
1445 		    || (children && fn->fn_flags & RTN_ROOT)
1446 #endif
1447 		    ) {
1448 			fn->leaf = fib6_find_prefix(net, fn);
1449 #if RT6_DEBUG >= 2
1450 			if (!fn->leaf) {
1451 				WARN_ON(!fn->leaf);
1452 				fn->leaf = net->ipv6.ip6_null_entry;
1453 			}
1454 #endif
1455 			atomic_inc(&fn->leaf->rt6i_ref);
1456 			return fn->parent;
1457 		}
1458 
1459 		pn = fn->parent;
1460 #ifdef CONFIG_IPV6_SUBTREES
1461 		if (FIB6_SUBTREE(pn) == fn) {
1462 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1463 			FIB6_SUBTREE(pn) = NULL;
1464 			nstate = FWS_L;
1465 		} else {
1466 			WARN_ON(fn->fn_flags & RTN_ROOT);
1467 #endif
1468 			if (pn->right == fn)
1469 				pn->right = child;
1470 			else if (pn->left == fn)
1471 				pn->left = child;
1472 #if RT6_DEBUG >= 2
1473 			else
1474 				WARN_ON(1);
1475 #endif
1476 			if (child)
1477 				child->parent = pn;
1478 			nstate = FWS_R;
1479 #ifdef CONFIG_IPV6_SUBTREES
1480 		}
1481 #endif
1482 
1483 		read_lock(&net->ipv6.fib6_walker_lock);
1484 		FOR_WALKERS(net, w) {
1485 			if (!child) {
1486 				if (w->root == fn) {
1487 					w->root = w->node = NULL;
1488 					RT6_TRACE("W %p adjusted by delroot 1\n", w);
1489 				} else if (w->node == fn) {
1490 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1491 					w->node = pn;
1492 					w->state = nstate;
1493 				}
1494 			} else {
1495 				if (w->root == fn) {
1496 					w->root = child;
1497 					RT6_TRACE("W %p adjusted by delroot 2\n", w);
1498 				}
1499 				if (w->node == fn) {
1500 					w->node = child;
1501 					if (children&2) {
1502 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1503 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1504 					} else {
1505 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1506 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1507 					}
1508 				}
1509 			}
1510 		}
1511 		read_unlock(&net->ipv6.fib6_walker_lock);
1512 
1513 		node_free(fn);
1514 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1515 			return pn;
1516 
1517 		rt6_release(pn->leaf);
1518 		pn->leaf = NULL;
1519 		fn = pn;
1520 	}
1521 }
1522 
1523 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1524 			   struct nl_info *info)
1525 {
1526 	struct fib6_walker *w;
1527 	struct rt6_info *rt = *rtp;
1528 	struct net *net = info->nl_net;
1529 
1530 	RT6_TRACE("fib6_del_route\n");
1531 
1532 	/* Unlink it */
1533 	*rtp = rt->dst.rt6_next;
1534 	rt->rt6i_node = NULL;
1535 	net->ipv6.rt6_stats->fib_rt_entries--;
1536 	net->ipv6.rt6_stats->fib_discarded_routes++;
1537 
1538 	/* Reset round-robin state, if necessary */
1539 	if (fn->rr_ptr == rt)
1540 		fn->rr_ptr = NULL;
1541 
1542 	/* Remove this entry from other siblings */
1543 	if (rt->rt6i_nsiblings) {
1544 		struct rt6_info *sibling, *next_sibling;
1545 
1546 		list_for_each_entry_safe(sibling, next_sibling,
1547 					 &rt->rt6i_siblings, rt6i_siblings)
1548 			sibling->rt6i_nsiblings--;
1549 		rt->rt6i_nsiblings = 0;
1550 		list_del_init(&rt->rt6i_siblings);
1551 	}
1552 
1553 	/* Adjust walkers */
1554 	read_lock(&net->ipv6.fib6_walker_lock);
1555 	FOR_WALKERS(net, w) {
1556 		if (w->state == FWS_C && w->leaf == rt) {
1557 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1558 			w->leaf = rt->dst.rt6_next;
1559 			if (!w->leaf)
1560 				w->state = FWS_U;
1561 		}
1562 	}
1563 	read_unlock(&net->ipv6.fib6_walker_lock);
1564 
1565 	rt->dst.rt6_next = NULL;
1566 
1567 	/* If it was last route, expunge its radix tree node */
1568 	if (!fn->leaf) {
1569 		fn->fn_flags &= ~RTN_RTINFO;
1570 		net->ipv6.rt6_stats->fib_route_nodes--;
1571 		fn = fib6_repair_tree(net, fn);
1572 	}
1573 
1574 	fib6_purge_rt(rt, fn, net);
1575 
1576 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt);
1577 	if (!info->skip_notify)
1578 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1579 	rt6_release(rt);
1580 }
1581 
1582 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1583 {
1584 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1585 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1586 	struct net *net = info->nl_net;
1587 	struct rt6_info **rtp;
1588 
1589 #if RT6_DEBUG >= 2
1590 	if (rt->dst.obsolete > 0) {
1591 		WARN_ON(fn);
1592 		return -ENOENT;
1593 	}
1594 #endif
1595 	if (!fn || rt == net->ipv6.ip6_null_entry)
1596 		return -ENOENT;
1597 
1598 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1599 
1600 	if (!(rt->rt6i_flags & RTF_CACHE)) {
1601 		struct fib6_node *pn = fn;
1602 #ifdef CONFIG_IPV6_SUBTREES
1603 		/* clones of this route might be in another subtree */
1604 		if (rt->rt6i_src.plen) {
1605 			while (!(pn->fn_flags & RTN_ROOT))
1606 				pn = pn->parent;
1607 			pn = pn->parent;
1608 		}
1609 #endif
1610 		fib6_prune_clones(info->nl_net, pn);
1611 	}
1612 
1613 	/*
1614 	 *	Walk the leaf entries looking for ourself
1615 	 */
1616 
1617 	for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1618 		if (*rtp == rt) {
1619 			fib6_del_route(fn, rtp, info);
1620 			return 0;
1621 		}
1622 	}
1623 	return -ENOENT;
1624 }
1625 
1626 /*
1627  *	Tree traversal function.
1628  *
1629  *	Certainly, it is not interrupt safe.
1630  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1631  *	It means, that we can modify tree during walking
1632  *	and use this function for garbage collection, clone pruning,
1633  *	cleaning tree when a device goes down etc. etc.
1634  *
1635  *	It guarantees that every node will be traversed,
1636  *	and that it will be traversed only once.
1637  *
1638  *	Callback function w->func may return:
1639  *	0 -> continue walking.
1640  *	positive value -> walking is suspended (used by tree dumps,
1641  *	and probably by gc, if it will be split to several slices)
1642  *	negative value -> terminate walking.
1643  *
1644  *	The function itself returns:
1645  *	0   -> walk is complete.
1646  *	>0  -> walk is incomplete (i.e. suspended)
1647  *	<0  -> walk is terminated by an error.
1648  */
1649 
1650 static int fib6_walk_continue(struct fib6_walker *w)
1651 {
1652 	struct fib6_node *fn, *pn;
1653 
1654 	for (;;) {
1655 		fn = w->node;
1656 		if (!fn)
1657 			return 0;
1658 
1659 		if (w->prune && fn != w->root &&
1660 		    fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1661 			w->state = FWS_C;
1662 			w->leaf = fn->leaf;
1663 		}
1664 		switch (w->state) {
1665 #ifdef CONFIG_IPV6_SUBTREES
1666 		case FWS_S:
1667 			if (FIB6_SUBTREE(fn)) {
1668 				w->node = FIB6_SUBTREE(fn);
1669 				continue;
1670 			}
1671 			w->state = FWS_L;
1672 #endif
1673 		case FWS_L:
1674 			if (fn->left) {
1675 				w->node = fn->left;
1676 				w->state = FWS_INIT;
1677 				continue;
1678 			}
1679 			w->state = FWS_R;
1680 		case FWS_R:
1681 			if (fn->right) {
1682 				w->node = fn->right;
1683 				w->state = FWS_INIT;
1684 				continue;
1685 			}
1686 			w->state = FWS_C;
1687 			w->leaf = fn->leaf;
1688 		case FWS_C:
1689 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1690 				int err;
1691 
1692 				if (w->skip) {
1693 					w->skip--;
1694 					goto skip;
1695 				}
1696 
1697 				err = w->func(w);
1698 				if (err)
1699 					return err;
1700 
1701 				w->count++;
1702 				continue;
1703 			}
1704 skip:
1705 			w->state = FWS_U;
1706 		case FWS_U:
1707 			if (fn == w->root)
1708 				return 0;
1709 			pn = fn->parent;
1710 			w->node = pn;
1711 #ifdef CONFIG_IPV6_SUBTREES
1712 			if (FIB6_SUBTREE(pn) == fn) {
1713 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1714 				w->state = FWS_L;
1715 				continue;
1716 			}
1717 #endif
1718 			if (pn->left == fn) {
1719 				w->state = FWS_R;
1720 				continue;
1721 			}
1722 			if (pn->right == fn) {
1723 				w->state = FWS_C;
1724 				w->leaf = w->node->leaf;
1725 				continue;
1726 			}
1727 #if RT6_DEBUG >= 2
1728 			WARN_ON(1);
1729 #endif
1730 		}
1731 	}
1732 }
1733 
1734 static int fib6_walk(struct net *net, struct fib6_walker *w)
1735 {
1736 	int res;
1737 
1738 	w->state = FWS_INIT;
1739 	w->node = w->root;
1740 
1741 	fib6_walker_link(net, w);
1742 	res = fib6_walk_continue(w);
1743 	if (res <= 0)
1744 		fib6_walker_unlink(net, w);
1745 	return res;
1746 }
1747 
1748 static int fib6_clean_node(struct fib6_walker *w)
1749 {
1750 	int res;
1751 	struct rt6_info *rt;
1752 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1753 	struct nl_info info = {
1754 		.nl_net = c->net,
1755 	};
1756 
1757 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1758 	    w->node->fn_sernum != c->sernum)
1759 		w->node->fn_sernum = c->sernum;
1760 
1761 	if (!c->func) {
1762 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1763 		w->leaf = NULL;
1764 		return 0;
1765 	}
1766 
1767 	for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1768 		res = c->func(rt, c->arg);
1769 		if (res < 0) {
1770 			w->leaf = rt;
1771 			res = fib6_del(rt, &info);
1772 			if (res) {
1773 #if RT6_DEBUG >= 2
1774 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1775 					 __func__, rt,
1776 					 rcu_access_pointer(rt->rt6i_node),
1777 					 res);
1778 #endif
1779 				continue;
1780 			}
1781 			return 0;
1782 		}
1783 		WARN_ON(res != 0);
1784 	}
1785 	w->leaf = rt;
1786 	return 0;
1787 }
1788 
1789 /*
1790  *	Convenient frontend to tree walker.
1791  *
1792  *	func is called on each route.
1793  *		It may return -1 -> delete this route.
1794  *		              0  -> continue walking
1795  *
1796  *	prune==1 -> only immediate children of node (certainly,
1797  *	ignoring pure split nodes) will be scanned.
1798  */
1799 
1800 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1801 			    int (*func)(struct rt6_info *, void *arg),
1802 			    bool prune, int sernum, void *arg)
1803 {
1804 	struct fib6_cleaner c;
1805 
1806 	c.w.root = root;
1807 	c.w.func = fib6_clean_node;
1808 	c.w.prune = prune;
1809 	c.w.count = 0;
1810 	c.w.skip = 0;
1811 	c.func = func;
1812 	c.sernum = sernum;
1813 	c.arg = arg;
1814 	c.net = net;
1815 
1816 	fib6_walk(net, &c.w);
1817 }
1818 
1819 static void __fib6_clean_all(struct net *net,
1820 			     int (*func)(struct rt6_info *, void *),
1821 			     int sernum, void *arg)
1822 {
1823 	struct fib6_table *table;
1824 	struct hlist_head *head;
1825 	unsigned int h;
1826 
1827 	rcu_read_lock();
1828 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1829 		head = &net->ipv6.fib_table_hash[h];
1830 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1831 			write_lock_bh(&table->tb6_lock);
1832 			fib6_clean_tree(net, &table->tb6_root,
1833 					func, false, sernum, arg);
1834 			write_unlock_bh(&table->tb6_lock);
1835 		}
1836 	}
1837 	rcu_read_unlock();
1838 }
1839 
1840 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1841 		    void *arg)
1842 {
1843 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1844 }
1845 
1846 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1847 {
1848 	if (rt->rt6i_flags & RTF_CACHE) {
1849 		RT6_TRACE("pruning clone %p\n", rt);
1850 		return -1;
1851 	}
1852 
1853 	return 0;
1854 }
1855 
1856 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1857 {
1858 	fib6_clean_tree(net, fn, fib6_prune_clone, true,
1859 			FIB6_NO_SERNUM_CHANGE, NULL);
1860 }
1861 
1862 static void fib6_flush_trees(struct net *net)
1863 {
1864 	int new_sernum = fib6_new_sernum(net);
1865 
1866 	__fib6_clean_all(net, NULL, new_sernum, NULL);
1867 }
1868 
1869 /*
1870  *	Garbage collection
1871  */
1872 
1873 struct fib6_gc_args
1874 {
1875 	int			timeout;
1876 	int			more;
1877 };
1878 
1879 static int fib6_age(struct rt6_info *rt, void *arg)
1880 {
1881 	struct fib6_gc_args *gc_args = arg;
1882 	unsigned long now = jiffies;
1883 
1884 	/*
1885 	 *	check addrconf expiration here.
1886 	 *	Routes are expired even if they are in use.
1887 	 *
1888 	 *	Also age clones. Note, that clones are aged out
1889 	 *	only if they are not in use now.
1890 	 */
1891 
1892 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1893 		if (time_after(now, rt->dst.expires)) {
1894 			RT6_TRACE("expiring %p\n", rt);
1895 			return -1;
1896 		}
1897 		gc_args->more++;
1898 	} else if (rt->rt6i_flags & RTF_CACHE) {
1899 		if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout))
1900 			rt->dst.obsolete = DST_OBSOLETE_KILL;
1901 		if (atomic_read(&rt->dst.__refcnt) == 1 &&
1902 		    rt->dst.obsolete == DST_OBSOLETE_KILL) {
1903 			RT6_TRACE("aging clone %p\n", rt);
1904 			return -1;
1905 		} else if (rt->rt6i_flags & RTF_GATEWAY) {
1906 			struct neighbour *neigh;
1907 			__u8 neigh_flags = 0;
1908 
1909 			neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1910 			if (neigh) {
1911 				neigh_flags = neigh->flags;
1912 				neigh_release(neigh);
1913 			}
1914 			if (!(neigh_flags & NTF_ROUTER)) {
1915 				RT6_TRACE("purging route %p via non-router but gateway\n",
1916 					  rt);
1917 				return -1;
1918 			}
1919 		}
1920 		gc_args->more++;
1921 	}
1922 
1923 	return 0;
1924 }
1925 
1926 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1927 {
1928 	struct fib6_gc_args gc_args;
1929 	unsigned long now;
1930 
1931 	if (force) {
1932 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
1933 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1934 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1935 		return;
1936 	}
1937 	gc_args.timeout = expires ? (int)expires :
1938 			  net->ipv6.sysctl.ip6_rt_gc_interval;
1939 	gc_args.more = 0;
1940 
1941 	fib6_clean_all(net, fib6_age, &gc_args);
1942 	now = jiffies;
1943 	net->ipv6.ip6_rt_last_gc = now;
1944 
1945 	if (gc_args.more)
1946 		mod_timer(&net->ipv6.ip6_fib_timer,
1947 			  round_jiffies(now
1948 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
1949 	else
1950 		del_timer(&net->ipv6.ip6_fib_timer);
1951 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1952 }
1953 
1954 static void fib6_gc_timer_cb(unsigned long arg)
1955 {
1956 	fib6_run_gc(0, (struct net *)arg, true);
1957 }
1958 
1959 static int __net_init fib6_net_init(struct net *net)
1960 {
1961 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1962 	int err;
1963 
1964 	err = fib6_notifier_init(net);
1965 	if (err)
1966 		return err;
1967 
1968 	spin_lock_init(&net->ipv6.fib6_gc_lock);
1969 	rwlock_init(&net->ipv6.fib6_walker_lock);
1970 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1971 	setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1972 
1973 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1974 	if (!net->ipv6.rt6_stats)
1975 		goto out_timer;
1976 
1977 	/* Avoid false sharing : Use at least a full cache line */
1978 	size = max_t(size_t, size, L1_CACHE_BYTES);
1979 
1980 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1981 	if (!net->ipv6.fib_table_hash)
1982 		goto out_rt6_stats;
1983 
1984 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1985 					  GFP_KERNEL);
1986 	if (!net->ipv6.fib6_main_tbl)
1987 		goto out_fib_table_hash;
1988 
1989 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1990 	net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1991 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1992 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1993 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1994 
1995 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1996 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1997 					   GFP_KERNEL);
1998 	if (!net->ipv6.fib6_local_tbl)
1999 		goto out_fib6_main_tbl;
2000 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2001 	net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
2002 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2003 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2004 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2005 #endif
2006 	fib6_tables_init(net);
2007 
2008 	return 0;
2009 
2010 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2011 out_fib6_main_tbl:
2012 	kfree(net->ipv6.fib6_main_tbl);
2013 #endif
2014 out_fib_table_hash:
2015 	kfree(net->ipv6.fib_table_hash);
2016 out_rt6_stats:
2017 	kfree(net->ipv6.rt6_stats);
2018 out_timer:
2019 	fib6_notifier_exit(net);
2020 	return -ENOMEM;
2021 }
2022 
2023 static void fib6_net_exit(struct net *net)
2024 {
2025 	rt6_ifdown(net, NULL);
2026 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2027 
2028 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2029 	inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
2030 	kfree(net->ipv6.fib6_local_tbl);
2031 #endif
2032 	inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
2033 	kfree(net->ipv6.fib6_main_tbl);
2034 	kfree(net->ipv6.fib_table_hash);
2035 	kfree(net->ipv6.rt6_stats);
2036 	fib6_notifier_exit(net);
2037 }
2038 
2039 static struct pernet_operations fib6_net_ops = {
2040 	.init = fib6_net_init,
2041 	.exit = fib6_net_exit,
2042 };
2043 
2044 int __init fib6_init(void)
2045 {
2046 	int ret = -ENOMEM;
2047 
2048 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2049 					   sizeof(struct fib6_node),
2050 					   0, SLAB_HWCACHE_ALIGN,
2051 					   NULL);
2052 	if (!fib6_node_kmem)
2053 		goto out;
2054 
2055 	ret = register_pernet_subsys(&fib6_net_ops);
2056 	if (ret)
2057 		goto out_kmem_cache_create;
2058 
2059 	ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
2060 			      0);
2061 	if (ret)
2062 		goto out_unregister_subsys;
2063 
2064 	__fib6_flush_trees = fib6_flush_trees;
2065 out:
2066 	return ret;
2067 
2068 out_unregister_subsys:
2069 	unregister_pernet_subsys(&fib6_net_ops);
2070 out_kmem_cache_create:
2071 	kmem_cache_destroy(fib6_node_kmem);
2072 	goto out;
2073 }
2074 
2075 void fib6_gc_cleanup(void)
2076 {
2077 	unregister_pernet_subsys(&fib6_net_ops);
2078 	kmem_cache_destroy(fib6_node_kmem);
2079 }
2080 
2081 #ifdef CONFIG_PROC_FS
2082 
2083 struct ipv6_route_iter {
2084 	struct seq_net_private p;
2085 	struct fib6_walker w;
2086 	loff_t skip;
2087 	struct fib6_table *tbl;
2088 	int sernum;
2089 };
2090 
2091 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2092 {
2093 	struct rt6_info *rt = v;
2094 	struct ipv6_route_iter *iter = seq->private;
2095 
2096 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2097 
2098 #ifdef CONFIG_IPV6_SUBTREES
2099 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2100 #else
2101 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2102 #endif
2103 	if (rt->rt6i_flags & RTF_GATEWAY)
2104 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2105 	else
2106 		seq_puts(seq, "00000000000000000000000000000000");
2107 
2108 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2109 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2110 		   rt->dst.__use, rt->rt6i_flags,
2111 		   rt->dst.dev ? rt->dst.dev->name : "");
2112 	iter->w.leaf = NULL;
2113 	return 0;
2114 }
2115 
2116 static int ipv6_route_yield(struct fib6_walker *w)
2117 {
2118 	struct ipv6_route_iter *iter = w->args;
2119 
2120 	if (!iter->skip)
2121 		return 1;
2122 
2123 	do {
2124 		iter->w.leaf = iter->w.leaf->dst.rt6_next;
2125 		iter->skip--;
2126 		if (!iter->skip && iter->w.leaf)
2127 			return 1;
2128 	} while (iter->w.leaf);
2129 
2130 	return 0;
2131 }
2132 
2133 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2134 				      struct net *net)
2135 {
2136 	memset(&iter->w, 0, sizeof(iter->w));
2137 	iter->w.func = ipv6_route_yield;
2138 	iter->w.root = &iter->tbl->tb6_root;
2139 	iter->w.state = FWS_INIT;
2140 	iter->w.node = iter->w.root;
2141 	iter->w.args = iter;
2142 	iter->sernum = iter->w.root->fn_sernum;
2143 	INIT_LIST_HEAD(&iter->w.lh);
2144 	fib6_walker_link(net, &iter->w);
2145 }
2146 
2147 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2148 						    struct net *net)
2149 {
2150 	unsigned int h;
2151 	struct hlist_node *node;
2152 
2153 	if (tbl) {
2154 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2155 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2156 	} else {
2157 		h = 0;
2158 		node = NULL;
2159 	}
2160 
2161 	while (!node && h < FIB6_TABLE_HASHSZ) {
2162 		node = rcu_dereference_bh(
2163 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2164 	}
2165 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2166 }
2167 
2168 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2169 {
2170 	if (iter->sernum != iter->w.root->fn_sernum) {
2171 		iter->sernum = iter->w.root->fn_sernum;
2172 		iter->w.state = FWS_INIT;
2173 		iter->w.node = iter->w.root;
2174 		WARN_ON(iter->w.skip);
2175 		iter->w.skip = iter->w.count;
2176 	}
2177 }
2178 
2179 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2180 {
2181 	int r;
2182 	struct rt6_info *n;
2183 	struct net *net = seq_file_net(seq);
2184 	struct ipv6_route_iter *iter = seq->private;
2185 
2186 	if (!v)
2187 		goto iter_table;
2188 
2189 	n = ((struct rt6_info *)v)->dst.rt6_next;
2190 	if (n) {
2191 		++*pos;
2192 		return n;
2193 	}
2194 
2195 iter_table:
2196 	ipv6_route_check_sernum(iter);
2197 	read_lock(&iter->tbl->tb6_lock);
2198 	r = fib6_walk_continue(&iter->w);
2199 	read_unlock(&iter->tbl->tb6_lock);
2200 	if (r > 0) {
2201 		if (v)
2202 			++*pos;
2203 		return iter->w.leaf;
2204 	} else if (r < 0) {
2205 		fib6_walker_unlink(net, &iter->w);
2206 		return NULL;
2207 	}
2208 	fib6_walker_unlink(net, &iter->w);
2209 
2210 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2211 	if (!iter->tbl)
2212 		return NULL;
2213 
2214 	ipv6_route_seq_setup_walk(iter, net);
2215 	goto iter_table;
2216 }
2217 
2218 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2219 	__acquires(RCU_BH)
2220 {
2221 	struct net *net = seq_file_net(seq);
2222 	struct ipv6_route_iter *iter = seq->private;
2223 
2224 	rcu_read_lock_bh();
2225 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2226 	iter->skip = *pos;
2227 
2228 	if (iter->tbl) {
2229 		ipv6_route_seq_setup_walk(iter, net);
2230 		return ipv6_route_seq_next(seq, NULL, pos);
2231 	} else {
2232 		return NULL;
2233 	}
2234 }
2235 
2236 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2237 {
2238 	struct fib6_walker *w = &iter->w;
2239 	return w->node && !(w->state == FWS_U && w->node == w->root);
2240 }
2241 
2242 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2243 	__releases(RCU_BH)
2244 {
2245 	struct net *net = seq_file_net(seq);
2246 	struct ipv6_route_iter *iter = seq->private;
2247 
2248 	if (ipv6_route_iter_active(iter))
2249 		fib6_walker_unlink(net, &iter->w);
2250 
2251 	rcu_read_unlock_bh();
2252 }
2253 
2254 static const struct seq_operations ipv6_route_seq_ops = {
2255 	.start	= ipv6_route_seq_start,
2256 	.next	= ipv6_route_seq_next,
2257 	.stop	= ipv6_route_seq_stop,
2258 	.show	= ipv6_route_seq_show
2259 };
2260 
2261 int ipv6_route_open(struct inode *inode, struct file *file)
2262 {
2263 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2264 			    sizeof(struct ipv6_route_iter));
2265 }
2266 
2267 #endif /* CONFIG_PROC_FS */
2268