xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision b75cc8f90f07342467b3bd51dbc0054f185032c9)
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 rt6_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 rt6_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 rt6_info *rt)
109 {
110 	struct net *net = dev_net(rt->dst.dev);
111 	struct fib6_node *fn;
112 
113 	fn = rcu_dereference_protected(rt->rt6i_node,
114 			lockdep_is_held(&rt->rt6i_table->tb6_lock));
115 	if (fn)
116 		fn->fn_sernum = fib6_new_sernum(net);
117 }
118 
119 /*
120  *	Auxiliary address test functions for the radix tree.
121  *
122  *	These assume a 32bit processor (although it will work on
123  *	64bit processors)
124  */
125 
126 /*
127  *	test bit
128  */
129 #if defined(__LITTLE_ENDIAN)
130 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
131 #else
132 # define BITOP_BE32_SWIZZLE	0
133 #endif
134 
135 static __be32 addr_bit_set(const void *token, int fn_bit)
136 {
137 	const __be32 *addr = token;
138 	/*
139 	 * Here,
140 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
141 	 * is optimized version of
142 	 *	htonl(1 << ((~fn_bit)&0x1F))
143 	 * See include/asm-generic/bitops/le.h.
144 	 */
145 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
146 	       addr[fn_bit >> 5];
147 }
148 
149 static struct fib6_node *node_alloc(struct net *net)
150 {
151 	struct fib6_node *fn;
152 
153 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
154 	if (fn)
155 		net->ipv6.rt6_stats->fib_nodes++;
156 
157 	return fn;
158 }
159 
160 static void node_free_immediate(struct net *net, struct fib6_node *fn)
161 {
162 	kmem_cache_free(fib6_node_kmem, fn);
163 	net->ipv6.rt6_stats->fib_nodes--;
164 }
165 
166 static void node_free_rcu(struct rcu_head *head)
167 {
168 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
169 
170 	kmem_cache_free(fib6_node_kmem, fn);
171 }
172 
173 static void node_free(struct net *net, struct fib6_node *fn)
174 {
175 	call_rcu(&fn->rcu, node_free_rcu);
176 	net->ipv6.rt6_stats->fib_nodes--;
177 }
178 
179 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
180 {
181 	int cpu;
182 
183 	if (!non_pcpu_rt->rt6i_pcpu)
184 		return;
185 
186 	for_each_possible_cpu(cpu) {
187 		struct rt6_info **ppcpu_rt;
188 		struct rt6_info *pcpu_rt;
189 
190 		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
191 		pcpu_rt = *ppcpu_rt;
192 		if (pcpu_rt) {
193 			dst_dev_put(&pcpu_rt->dst);
194 			dst_release(&pcpu_rt->dst);
195 			*ppcpu_rt = NULL;
196 		}
197 	}
198 }
199 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
200 
201 static void fib6_free_table(struct fib6_table *table)
202 {
203 	inetpeer_invalidate_tree(&table->tb6_peers);
204 	kfree(table);
205 }
206 
207 static void fib6_link_table(struct net *net, struct fib6_table *tb)
208 {
209 	unsigned int h;
210 
211 	/*
212 	 * Initialize table lock at a single place to give lockdep a key,
213 	 * tables aren't visible prior to being linked to the list.
214 	 */
215 	spin_lock_init(&tb->tb6_lock);
216 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
217 
218 	/*
219 	 * No protection necessary, this is the only list mutatation
220 	 * operation, tables never disappear once they exist.
221 	 */
222 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
223 }
224 
225 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
226 
227 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
228 {
229 	struct fib6_table *table;
230 
231 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
232 	if (table) {
233 		table->tb6_id = id;
234 		rcu_assign_pointer(table->tb6_root.leaf,
235 				   net->ipv6.ip6_null_entry);
236 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
237 		inet_peer_base_init(&table->tb6_peers);
238 	}
239 
240 	return table;
241 }
242 
243 struct fib6_table *fib6_new_table(struct net *net, u32 id)
244 {
245 	struct fib6_table *tb;
246 
247 	if (id == 0)
248 		id = RT6_TABLE_MAIN;
249 	tb = fib6_get_table(net, id);
250 	if (tb)
251 		return tb;
252 
253 	tb = fib6_alloc_table(net, id);
254 	if (tb)
255 		fib6_link_table(net, tb);
256 
257 	return tb;
258 }
259 EXPORT_SYMBOL_GPL(fib6_new_table);
260 
261 struct fib6_table *fib6_get_table(struct net *net, u32 id)
262 {
263 	struct fib6_table *tb;
264 	struct hlist_head *head;
265 	unsigned int h;
266 
267 	if (id == 0)
268 		id = RT6_TABLE_MAIN;
269 	h = id & (FIB6_TABLE_HASHSZ - 1);
270 	rcu_read_lock();
271 	head = &net->ipv6.fib_table_hash[h];
272 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
273 		if (tb->tb6_id == id) {
274 			rcu_read_unlock();
275 			return tb;
276 		}
277 	}
278 	rcu_read_unlock();
279 
280 	return NULL;
281 }
282 EXPORT_SYMBOL_GPL(fib6_get_table);
283 
284 static void __net_init fib6_tables_init(struct net *net)
285 {
286 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
287 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
288 }
289 #else
290 
291 struct fib6_table *fib6_new_table(struct net *net, u32 id)
292 {
293 	return fib6_get_table(net, id);
294 }
295 
296 struct fib6_table *fib6_get_table(struct net *net, u32 id)
297 {
298 	  return net->ipv6.fib6_main_tbl;
299 }
300 
301 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
302 				   const struct sk_buff *skb,
303 				   int flags, pol_lookup_t lookup)
304 {
305 	struct rt6_info *rt;
306 
307 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
308 	if (rt->dst.error == -EAGAIN) {
309 		ip6_rt_put(rt);
310 		rt = net->ipv6.ip6_null_entry;
311 		dst_hold(&rt->dst);
312 	}
313 
314 	return &rt->dst;
315 }
316 
317 static void __net_init fib6_tables_init(struct net *net)
318 {
319 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
320 }
321 
322 #endif
323 
324 unsigned int fib6_tables_seq_read(struct net *net)
325 {
326 	unsigned int h, fib_seq = 0;
327 
328 	rcu_read_lock();
329 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
330 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
331 		struct fib6_table *tb;
332 
333 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
334 			fib_seq += tb->fib_seq;
335 	}
336 	rcu_read_unlock();
337 
338 	return fib_seq;
339 }
340 
341 static int call_fib6_entry_notifier(struct notifier_block *nb, 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 	return call_fib6_notifier(nb, net, event_type, &info.info);
350 }
351 
352 static int call_fib6_entry_notifiers(struct net *net,
353 				     enum fib_event_type event_type,
354 				     struct rt6_info *rt,
355 				     struct netlink_ext_ack *extack)
356 {
357 	struct fib6_entry_notifier_info info = {
358 		.info.extack = extack,
359 		.rt = rt,
360 	};
361 
362 	rt->rt6i_table->fib_seq++;
363 	return call_fib6_notifiers(net, event_type, &info.info);
364 }
365 
366 struct fib6_dump_arg {
367 	struct net *net;
368 	struct notifier_block *nb;
369 };
370 
371 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
372 {
373 	if (rt == arg->net->ipv6.ip6_null_entry)
374 		return;
375 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
376 }
377 
378 static int fib6_node_dump(struct fib6_walker *w)
379 {
380 	struct rt6_info *rt;
381 
382 	for_each_fib6_walker_rt(w)
383 		fib6_rt_dump(rt, w->args);
384 	w->leaf = NULL;
385 	return 0;
386 }
387 
388 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
389 			    struct fib6_walker *w)
390 {
391 	w->root = &tb->tb6_root;
392 	spin_lock_bh(&tb->tb6_lock);
393 	fib6_walk(net, w);
394 	spin_unlock_bh(&tb->tb6_lock);
395 }
396 
397 /* Called with rcu_read_lock() */
398 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
399 {
400 	struct fib6_dump_arg arg;
401 	struct fib6_walker *w;
402 	unsigned int h;
403 
404 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
405 	if (!w)
406 		return -ENOMEM;
407 
408 	w->func = fib6_node_dump;
409 	arg.net = net;
410 	arg.nb = nb;
411 	w->args = &arg;
412 
413 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
414 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
415 		struct fib6_table *tb;
416 
417 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
418 			fib6_table_dump(net, tb, w);
419 	}
420 
421 	kfree(w);
422 
423 	return 0;
424 }
425 
426 static int fib6_dump_node(struct fib6_walker *w)
427 {
428 	int res;
429 	struct rt6_info *rt;
430 
431 	for_each_fib6_walker_rt(w) {
432 		res = rt6_dump_route(rt, w->args);
433 		if (res < 0) {
434 			/* Frame is full, suspend walking */
435 			w->leaf = rt;
436 			return 1;
437 		}
438 
439 		/* Multipath routes are dumped in one route with the
440 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
441 		 * last sibling of this route (no need to dump the
442 		 * sibling routes again)
443 		 */
444 		if (rt->rt6i_nsiblings)
445 			rt = list_last_entry(&rt->rt6i_siblings,
446 					     struct rt6_info,
447 					     rt6i_siblings);
448 	}
449 	w->leaf = NULL;
450 	return 0;
451 }
452 
453 static void fib6_dump_end(struct netlink_callback *cb)
454 {
455 	struct net *net = sock_net(cb->skb->sk);
456 	struct fib6_walker *w = (void *)cb->args[2];
457 
458 	if (w) {
459 		if (cb->args[4]) {
460 			cb->args[4] = 0;
461 			fib6_walker_unlink(net, w);
462 		}
463 		cb->args[2] = 0;
464 		kfree(w);
465 	}
466 	cb->done = (void *)cb->args[3];
467 	cb->args[1] = 3;
468 }
469 
470 static int fib6_dump_done(struct netlink_callback *cb)
471 {
472 	fib6_dump_end(cb);
473 	return cb->done ? cb->done(cb) : 0;
474 }
475 
476 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
477 			   struct netlink_callback *cb)
478 {
479 	struct net *net = sock_net(skb->sk);
480 	struct fib6_walker *w;
481 	int res;
482 
483 	w = (void *)cb->args[2];
484 	w->root = &table->tb6_root;
485 
486 	if (cb->args[4] == 0) {
487 		w->count = 0;
488 		w->skip = 0;
489 
490 		spin_lock_bh(&table->tb6_lock);
491 		res = fib6_walk(net, w);
492 		spin_unlock_bh(&table->tb6_lock);
493 		if (res > 0) {
494 			cb->args[4] = 1;
495 			cb->args[5] = w->root->fn_sernum;
496 		}
497 	} else {
498 		if (cb->args[5] != w->root->fn_sernum) {
499 			/* Begin at the root if the tree changed */
500 			cb->args[5] = w->root->fn_sernum;
501 			w->state = FWS_INIT;
502 			w->node = w->root;
503 			w->skip = w->count;
504 		} else
505 			w->skip = 0;
506 
507 		spin_lock_bh(&table->tb6_lock);
508 		res = fib6_walk_continue(w);
509 		spin_unlock_bh(&table->tb6_lock);
510 		if (res <= 0) {
511 			fib6_walker_unlink(net, w);
512 			cb->args[4] = 0;
513 		}
514 	}
515 
516 	return res;
517 }
518 
519 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
520 {
521 	struct net *net = sock_net(skb->sk);
522 	unsigned int h, s_h;
523 	unsigned int e = 0, s_e;
524 	struct rt6_rtnl_dump_arg arg;
525 	struct fib6_walker *w;
526 	struct fib6_table *tb;
527 	struct hlist_head *head;
528 	int res = 0;
529 
530 	s_h = cb->args[0];
531 	s_e = cb->args[1];
532 
533 	w = (void *)cb->args[2];
534 	if (!w) {
535 		/* New dump:
536 		 *
537 		 * 1. hook callback destructor.
538 		 */
539 		cb->args[3] = (long)cb->done;
540 		cb->done = fib6_dump_done;
541 
542 		/*
543 		 * 2. allocate and initialize walker.
544 		 */
545 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
546 		if (!w)
547 			return -ENOMEM;
548 		w->func = fib6_dump_node;
549 		cb->args[2] = (long)w;
550 	}
551 
552 	arg.skb = skb;
553 	arg.cb = cb;
554 	arg.net = net;
555 	w->args = &arg;
556 
557 	rcu_read_lock();
558 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
559 		e = 0;
560 		head = &net->ipv6.fib_table_hash[h];
561 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
562 			if (e < s_e)
563 				goto next;
564 			res = fib6_dump_table(tb, skb, cb);
565 			if (res != 0)
566 				goto out;
567 next:
568 			e++;
569 		}
570 	}
571 out:
572 	rcu_read_unlock();
573 	cb->args[1] = e;
574 	cb->args[0] = h;
575 
576 	res = res < 0 ? res : skb->len;
577 	if (res <= 0)
578 		fib6_dump_end(cb);
579 	return res;
580 }
581 
582 /*
583  *	Routing Table
584  *
585  *	return the appropriate node for a routing tree "add" operation
586  *	by either creating and inserting or by returning an existing
587  *	node.
588  */
589 
590 static struct fib6_node *fib6_add_1(struct net *net,
591 				    struct fib6_table *table,
592 				    struct fib6_node *root,
593 				    struct in6_addr *addr, int plen,
594 				    int offset, int allow_create,
595 				    int replace_required,
596 				    struct netlink_ext_ack *extack)
597 {
598 	struct fib6_node *fn, *in, *ln;
599 	struct fib6_node *pn = NULL;
600 	struct rt6key *key;
601 	int	bit;
602 	__be32	dir = 0;
603 
604 	RT6_TRACE("fib6_add_1\n");
605 
606 	/* insert node in tree */
607 
608 	fn = root;
609 
610 	do {
611 		struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
612 					    lockdep_is_held(&table->tb6_lock));
613 		key = (struct rt6key *)((u8 *)leaf + offset);
614 
615 		/*
616 		 *	Prefix match
617 		 */
618 		if (plen < fn->fn_bit ||
619 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
620 			if (!allow_create) {
621 				if (replace_required) {
622 					NL_SET_ERR_MSG(extack,
623 						       "Can not replace route - no match found");
624 					pr_warn("Can't replace route, no match found\n");
625 					return ERR_PTR(-ENOENT);
626 				}
627 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
628 			}
629 			goto insert_above;
630 		}
631 
632 		/*
633 		 *	Exact match ?
634 		 */
635 
636 		if (plen == fn->fn_bit) {
637 			/* clean up an intermediate node */
638 			if (!(fn->fn_flags & RTN_RTINFO)) {
639 				RCU_INIT_POINTER(fn->leaf, NULL);
640 				rt6_release(leaf);
641 			/* remove null_entry in the root node */
642 			} else if (fn->fn_flags & RTN_TL_ROOT &&
643 				   rcu_access_pointer(fn->leaf) ==
644 				   net->ipv6.ip6_null_entry) {
645 				RCU_INIT_POINTER(fn->leaf, NULL);
646 			}
647 
648 			return fn;
649 		}
650 
651 		/*
652 		 *	We have more bits to go
653 		 */
654 
655 		/* Try to walk down on tree. */
656 		dir = addr_bit_set(addr, fn->fn_bit);
657 		pn = fn;
658 		fn = dir ?
659 		     rcu_dereference_protected(fn->right,
660 					lockdep_is_held(&table->tb6_lock)) :
661 		     rcu_dereference_protected(fn->left,
662 					lockdep_is_held(&table->tb6_lock));
663 	} while (fn);
664 
665 	if (!allow_create) {
666 		/* We should not create new node because
667 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
668 		 * I assume it is safe to require NLM_F_CREATE when
669 		 * REPLACE flag is used! Later we may want to remove the
670 		 * check for replace_required, because according
671 		 * to netlink specification, NLM_F_CREATE
672 		 * MUST be specified if new route is created.
673 		 * That would keep IPv6 consistent with IPv4
674 		 */
675 		if (replace_required) {
676 			NL_SET_ERR_MSG(extack,
677 				       "Can not replace route - no match found");
678 			pr_warn("Can't replace route, no match found\n");
679 			return ERR_PTR(-ENOENT);
680 		}
681 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
682 	}
683 	/*
684 	 *	We walked to the bottom of tree.
685 	 *	Create new leaf node without children.
686 	 */
687 
688 	ln = node_alloc(net);
689 
690 	if (!ln)
691 		return ERR_PTR(-ENOMEM);
692 	ln->fn_bit = plen;
693 	RCU_INIT_POINTER(ln->parent, pn);
694 
695 	if (dir)
696 		rcu_assign_pointer(pn->right, ln);
697 	else
698 		rcu_assign_pointer(pn->left, ln);
699 
700 	return ln;
701 
702 
703 insert_above:
704 	/*
705 	 * split since we don't have a common prefix anymore or
706 	 * we have a less significant route.
707 	 * we've to insert an intermediate node on the list
708 	 * this new node will point to the one we need to create
709 	 * and the current
710 	 */
711 
712 	pn = rcu_dereference_protected(fn->parent,
713 				       lockdep_is_held(&table->tb6_lock));
714 
715 	/* find 1st bit in difference between the 2 addrs.
716 
717 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
718 	   but if it is >= plen, the value is ignored in any case.
719 	 */
720 
721 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
722 
723 	/*
724 	 *		(intermediate)[in]
725 	 *	          /	   \
726 	 *	(new leaf node)[ln] (old node)[fn]
727 	 */
728 	if (plen > bit) {
729 		in = node_alloc(net);
730 		ln = node_alloc(net);
731 
732 		if (!in || !ln) {
733 			if (in)
734 				node_free_immediate(net, in);
735 			if (ln)
736 				node_free_immediate(net, ln);
737 			return ERR_PTR(-ENOMEM);
738 		}
739 
740 		/*
741 		 * new intermediate node.
742 		 * RTN_RTINFO will
743 		 * be off since that an address that chooses one of
744 		 * the branches would not match less specific routes
745 		 * in the other branch
746 		 */
747 
748 		in->fn_bit = bit;
749 
750 		RCU_INIT_POINTER(in->parent, pn);
751 		in->leaf = fn->leaf;
752 		atomic_inc(&rcu_dereference_protected(in->leaf,
753 				lockdep_is_held(&table->tb6_lock))->rt6i_ref);
754 
755 		/* update parent pointer */
756 		if (dir)
757 			rcu_assign_pointer(pn->right, in);
758 		else
759 			rcu_assign_pointer(pn->left, in);
760 
761 		ln->fn_bit = plen;
762 
763 		RCU_INIT_POINTER(ln->parent, in);
764 		rcu_assign_pointer(fn->parent, in);
765 
766 		if (addr_bit_set(addr, bit)) {
767 			rcu_assign_pointer(in->right, ln);
768 			rcu_assign_pointer(in->left, fn);
769 		} else {
770 			rcu_assign_pointer(in->left, ln);
771 			rcu_assign_pointer(in->right, fn);
772 		}
773 	} else { /* plen <= bit */
774 
775 		/*
776 		 *		(new leaf node)[ln]
777 		 *	          /	   \
778 		 *	     (old node)[fn] NULL
779 		 */
780 
781 		ln = node_alloc(net);
782 
783 		if (!ln)
784 			return ERR_PTR(-ENOMEM);
785 
786 		ln->fn_bit = plen;
787 
788 		RCU_INIT_POINTER(ln->parent, pn);
789 
790 		if (addr_bit_set(&key->addr, plen))
791 			RCU_INIT_POINTER(ln->right, fn);
792 		else
793 			RCU_INIT_POINTER(ln->left, fn);
794 
795 		rcu_assign_pointer(fn->parent, ln);
796 
797 		if (dir)
798 			rcu_assign_pointer(pn->right, ln);
799 		else
800 			rcu_assign_pointer(pn->left, ln);
801 	}
802 	return ln;
803 }
804 
805 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
806 {
807 	int i;
808 
809 	for (i = 0; i < RTAX_MAX; i++) {
810 		if (test_bit(i, mxc->mx_valid))
811 			mp[i] = mxc->mx[i];
812 	}
813 }
814 
815 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
816 {
817 	if (!mxc->mx)
818 		return 0;
819 
820 	if (dst->flags & DST_HOST) {
821 		u32 *mp = dst_metrics_write_ptr(dst);
822 
823 		if (unlikely(!mp))
824 			return -ENOMEM;
825 
826 		fib6_copy_metrics(mp, mxc);
827 	} else {
828 		dst_init_metrics(dst, mxc->mx, false);
829 
830 		/* We've stolen mx now. */
831 		mxc->mx = NULL;
832 	}
833 
834 	return 0;
835 }
836 
837 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
838 			  struct net *net)
839 {
840 	struct fib6_table *table = rt->rt6i_table;
841 
842 	if (atomic_read(&rt->rt6i_ref) != 1) {
843 		/* This route is used as dummy address holder in some split
844 		 * nodes. It is not leaked, but it still holds other resources,
845 		 * which must be released in time. So, scan ascendant nodes
846 		 * and replace dummy references to this route with references
847 		 * to still alive ones.
848 		 */
849 		while (fn) {
850 			struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
851 					    lockdep_is_held(&table->tb6_lock));
852 			struct rt6_info *new_leaf;
853 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
854 				new_leaf = fib6_find_prefix(net, table, fn);
855 				atomic_inc(&new_leaf->rt6i_ref);
856 				rcu_assign_pointer(fn->leaf, new_leaf);
857 				rt6_release(rt);
858 			}
859 			fn = rcu_dereference_protected(fn->parent,
860 				    lockdep_is_held(&table->tb6_lock));
861 		}
862 	}
863 }
864 
865 /*
866  *	Insert routing information in a node.
867  */
868 
869 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
870 			    struct nl_info *info, struct mx6_config *mxc,
871 			    struct netlink_ext_ack *extack)
872 {
873 	struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
874 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
875 	struct rt6_info *iter = NULL;
876 	struct rt6_info __rcu **ins;
877 	struct rt6_info __rcu **fallback_ins = NULL;
878 	int replace = (info->nlh &&
879 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
880 	int add = (!info->nlh ||
881 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
882 	int found = 0;
883 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
884 	u16 nlflags = NLM_F_EXCL;
885 	int err;
886 
887 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
888 		nlflags |= NLM_F_APPEND;
889 
890 	ins = &fn->leaf;
891 
892 	for (iter = leaf; iter;
893 	     iter = rcu_dereference_protected(iter->rt6_next,
894 				lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
895 		/*
896 		 *	Search for duplicates
897 		 */
898 
899 		if (iter->rt6i_metric == rt->rt6i_metric) {
900 			/*
901 			 *	Same priority level
902 			 */
903 			if (info->nlh &&
904 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
905 				return -EEXIST;
906 
907 			nlflags &= ~NLM_F_EXCL;
908 			if (replace) {
909 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
910 					found++;
911 					break;
912 				}
913 				if (rt_can_ecmp)
914 					fallback_ins = fallback_ins ?: ins;
915 				goto next_iter;
916 			}
917 
918 			if (rt6_duplicate_nexthop(iter, rt)) {
919 				if (rt->rt6i_nsiblings)
920 					rt->rt6i_nsiblings = 0;
921 				if (!(iter->rt6i_flags & RTF_EXPIRES))
922 					return -EEXIST;
923 				if (!(rt->rt6i_flags & RTF_EXPIRES))
924 					rt6_clean_expires(iter);
925 				else
926 					rt6_set_expires(iter, rt->dst.expires);
927 				iter->rt6i_pmtu = rt->rt6i_pmtu;
928 				return -EEXIST;
929 			}
930 			/* If we have the same destination and the same metric,
931 			 * but not the same gateway, then the route we try to
932 			 * add is sibling to this route, increment our counter
933 			 * of siblings, and later we will add our route to the
934 			 * list.
935 			 * Only static routes (which don't have flag
936 			 * RTF_EXPIRES) are used for ECMPv6.
937 			 *
938 			 * To avoid long list, we only had siblings if the
939 			 * route have a gateway.
940 			 */
941 			if (rt_can_ecmp &&
942 			    rt6_qualify_for_ecmp(iter))
943 				rt->rt6i_nsiblings++;
944 		}
945 
946 		if (iter->rt6i_metric > rt->rt6i_metric)
947 			break;
948 
949 next_iter:
950 		ins = &iter->rt6_next;
951 	}
952 
953 	if (fallback_ins && !found) {
954 		/* No ECMP-able route found, replace first non-ECMP one */
955 		ins = fallback_ins;
956 		iter = rcu_dereference_protected(*ins,
957 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
958 		found++;
959 	}
960 
961 	/* Reset round-robin state, if necessary */
962 	if (ins == &fn->leaf)
963 		fn->rr_ptr = NULL;
964 
965 	/* Link this route to others same route. */
966 	if (rt->rt6i_nsiblings) {
967 		unsigned int rt6i_nsiblings;
968 		struct rt6_info *sibling, *temp_sibling;
969 
970 		/* Find the first route that have the same metric */
971 		sibling = leaf;
972 		while (sibling) {
973 			if (sibling->rt6i_metric == rt->rt6i_metric &&
974 			    rt6_qualify_for_ecmp(sibling)) {
975 				list_add_tail(&rt->rt6i_siblings,
976 					      &sibling->rt6i_siblings);
977 				break;
978 			}
979 			sibling = rcu_dereference_protected(sibling->rt6_next,
980 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
981 		}
982 		/* For each sibling in the list, increment the counter of
983 		 * siblings. BUG() if counters does not match, list of siblings
984 		 * is broken!
985 		 */
986 		rt6i_nsiblings = 0;
987 		list_for_each_entry_safe(sibling, temp_sibling,
988 					 &rt->rt6i_siblings, rt6i_siblings) {
989 			sibling->rt6i_nsiblings++;
990 			BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
991 			rt6i_nsiblings++;
992 		}
993 		BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
994 		rt6_multipath_rebalance(temp_sibling);
995 	}
996 
997 	/*
998 	 *	insert node
999 	 */
1000 	if (!replace) {
1001 		if (!add)
1002 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1003 
1004 add:
1005 		nlflags |= NLM_F_CREATE;
1006 		err = fib6_commit_metrics(&rt->dst, mxc);
1007 		if (err)
1008 			return err;
1009 
1010 		rcu_assign_pointer(rt->rt6_next, iter);
1011 		atomic_inc(&rt->rt6i_ref);
1012 		rcu_assign_pointer(rt->rt6i_node, fn);
1013 		rcu_assign_pointer(*ins, rt);
1014 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
1015 					  rt, extack);
1016 		if (!info->skip_notify)
1017 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1018 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1019 
1020 		if (!(fn->fn_flags & RTN_RTINFO)) {
1021 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1022 			fn->fn_flags |= RTN_RTINFO;
1023 		}
1024 
1025 	} else {
1026 		int nsiblings;
1027 
1028 		if (!found) {
1029 			if (add)
1030 				goto add;
1031 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1032 			return -ENOENT;
1033 		}
1034 
1035 		err = fib6_commit_metrics(&rt->dst, mxc);
1036 		if (err)
1037 			return err;
1038 
1039 		atomic_inc(&rt->rt6i_ref);
1040 		rcu_assign_pointer(rt->rt6i_node, fn);
1041 		rt->rt6_next = iter->rt6_next;
1042 		rcu_assign_pointer(*ins, rt);
1043 		call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1044 					  rt, extack);
1045 		if (!info->skip_notify)
1046 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1047 		if (!(fn->fn_flags & RTN_RTINFO)) {
1048 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1049 			fn->fn_flags |= RTN_RTINFO;
1050 		}
1051 		nsiblings = iter->rt6i_nsiblings;
1052 		iter->rt6i_node = NULL;
1053 		fib6_purge_rt(iter, fn, info->nl_net);
1054 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1055 			fn->rr_ptr = NULL;
1056 		rt6_release(iter);
1057 
1058 		if (nsiblings) {
1059 			/* Replacing an ECMP route, remove all siblings */
1060 			ins = &rt->rt6_next;
1061 			iter = rcu_dereference_protected(*ins,
1062 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1063 			while (iter) {
1064 				if (iter->rt6i_metric > rt->rt6i_metric)
1065 					break;
1066 				if (rt6_qualify_for_ecmp(iter)) {
1067 					*ins = iter->rt6_next;
1068 					iter->rt6i_node = NULL;
1069 					fib6_purge_rt(iter, fn, info->nl_net);
1070 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1071 						fn->rr_ptr = NULL;
1072 					rt6_release(iter);
1073 					nsiblings--;
1074 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1075 				} else {
1076 					ins = &iter->rt6_next;
1077 				}
1078 				iter = rcu_dereference_protected(*ins,
1079 					lockdep_is_held(&rt->rt6i_table->tb6_lock));
1080 			}
1081 			WARN_ON(nsiblings != 0);
1082 		}
1083 	}
1084 
1085 	return 0;
1086 }
1087 
1088 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1089 {
1090 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1091 	    (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1092 		mod_timer(&net->ipv6.ip6_fib_timer,
1093 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1094 }
1095 
1096 void fib6_force_start_gc(struct net *net)
1097 {
1098 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1099 		mod_timer(&net->ipv6.ip6_fib_timer,
1100 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1101 }
1102 
1103 static void __fib6_update_sernum_upto_root(struct rt6_info *rt,
1104 					   int sernum)
1105 {
1106 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1107 				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1108 
1109 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1110 	smp_wmb();
1111 	while (fn) {
1112 		fn->fn_sernum = sernum;
1113 		fn = rcu_dereference_protected(fn->parent,
1114 				lockdep_is_held(&rt->rt6i_table->tb6_lock));
1115 	}
1116 }
1117 
1118 void fib6_update_sernum_upto_root(struct net *net, struct rt6_info *rt)
1119 {
1120 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1121 }
1122 
1123 /*
1124  *	Add routing information to the routing tree.
1125  *	<destination addr>/<source addr>
1126  *	with source addr info in sub-trees
1127  *	Need to own table->tb6_lock
1128  */
1129 
1130 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1131 	     struct nl_info *info, struct mx6_config *mxc,
1132 	     struct netlink_ext_ack *extack)
1133 {
1134 	struct fib6_table *table = rt->rt6i_table;
1135 	struct fib6_node *fn, *pn = NULL;
1136 	int err = -ENOMEM;
1137 	int allow_create = 1;
1138 	int replace_required = 0;
1139 	int sernum = fib6_new_sernum(info->nl_net);
1140 
1141 	if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1142 		return -EINVAL;
1143 	if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1144 		return -EINVAL;
1145 
1146 	if (info->nlh) {
1147 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1148 			allow_create = 0;
1149 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1150 			replace_required = 1;
1151 	}
1152 	if (!allow_create && !replace_required)
1153 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1154 
1155 	fn = fib6_add_1(info->nl_net, table, root,
1156 			&rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1157 			offsetof(struct rt6_info, rt6i_dst), allow_create,
1158 			replace_required, extack);
1159 	if (IS_ERR(fn)) {
1160 		err = PTR_ERR(fn);
1161 		fn = NULL;
1162 		goto out;
1163 	}
1164 
1165 	pn = fn;
1166 
1167 #ifdef CONFIG_IPV6_SUBTREES
1168 	if (rt->rt6i_src.plen) {
1169 		struct fib6_node *sn;
1170 
1171 		if (!rcu_access_pointer(fn->subtree)) {
1172 			struct fib6_node *sfn;
1173 
1174 			/*
1175 			 * Create subtree.
1176 			 *
1177 			 *		fn[main tree]
1178 			 *		|
1179 			 *		sfn[subtree root]
1180 			 *		   \
1181 			 *		    sn[new leaf node]
1182 			 */
1183 
1184 			/* Create subtree root node */
1185 			sfn = node_alloc(info->nl_net);
1186 			if (!sfn)
1187 				goto failure;
1188 
1189 			atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1190 			rcu_assign_pointer(sfn->leaf,
1191 					   info->nl_net->ipv6.ip6_null_entry);
1192 			sfn->fn_flags = RTN_ROOT;
1193 
1194 			/* Now add the first leaf node to new subtree */
1195 
1196 			sn = fib6_add_1(info->nl_net, table, sfn,
1197 					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1198 					offsetof(struct rt6_info, rt6i_src),
1199 					allow_create, replace_required, extack);
1200 
1201 			if (IS_ERR(sn)) {
1202 				/* If it is failed, discard just allocated
1203 				   root, and then (in failure) stale node
1204 				   in main tree.
1205 				 */
1206 				node_free_immediate(info->nl_net, sfn);
1207 				err = PTR_ERR(sn);
1208 				goto failure;
1209 			}
1210 
1211 			/* Now link new subtree to main tree */
1212 			rcu_assign_pointer(sfn->parent, fn);
1213 			rcu_assign_pointer(fn->subtree, sfn);
1214 		} else {
1215 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1216 					&rt->rt6i_src.addr, rt->rt6i_src.plen,
1217 					offsetof(struct rt6_info, rt6i_src),
1218 					allow_create, replace_required, extack);
1219 
1220 			if (IS_ERR(sn)) {
1221 				err = PTR_ERR(sn);
1222 				goto failure;
1223 			}
1224 		}
1225 
1226 		if (!rcu_access_pointer(fn->leaf)) {
1227 			if (fn->fn_flags & RTN_TL_ROOT) {
1228 				/* put back null_entry for root node */
1229 				rcu_assign_pointer(fn->leaf,
1230 					    info->nl_net->ipv6.ip6_null_entry);
1231 			} else {
1232 				atomic_inc(&rt->rt6i_ref);
1233 				rcu_assign_pointer(fn->leaf, rt);
1234 			}
1235 		}
1236 		fn = sn;
1237 	}
1238 #endif
1239 
1240 	err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1241 	if (!err) {
1242 		__fib6_update_sernum_upto_root(rt, sernum);
1243 		fib6_start_gc(info->nl_net, rt);
1244 	}
1245 
1246 out:
1247 	if (err) {
1248 #ifdef CONFIG_IPV6_SUBTREES
1249 		/*
1250 		 * If fib6_add_1 has cleared the old leaf pointer in the
1251 		 * super-tree leaf node we have to find a new one for it.
1252 		 */
1253 		if (pn != fn) {
1254 			struct rt6_info *pn_leaf =
1255 				rcu_dereference_protected(pn->leaf,
1256 				    lockdep_is_held(&table->tb6_lock));
1257 			if (pn_leaf == rt) {
1258 				pn_leaf = NULL;
1259 				RCU_INIT_POINTER(pn->leaf, NULL);
1260 				atomic_dec(&rt->rt6i_ref);
1261 			}
1262 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1263 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1264 							   pn);
1265 #if RT6_DEBUG >= 2
1266 				if (!pn_leaf) {
1267 					WARN_ON(!pn_leaf);
1268 					pn_leaf =
1269 					    info->nl_net->ipv6.ip6_null_entry;
1270 				}
1271 #endif
1272 				atomic_inc(&pn_leaf->rt6i_ref);
1273 				rcu_assign_pointer(pn->leaf, pn_leaf);
1274 			}
1275 		}
1276 #endif
1277 		goto failure;
1278 	}
1279 	return err;
1280 
1281 failure:
1282 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1283 	 * 1. fn is an intermediate node and we failed to add the new
1284 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1285 	 * failure case.
1286 	 * 2. fn is the root node in the table and we fail to add the first
1287 	 * default route to it.
1288 	 */
1289 	if (fn &&
1290 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1291 	     (fn->fn_flags & RTN_TL_ROOT &&
1292 	      !rcu_access_pointer(fn->leaf))))
1293 		fib6_repair_tree(info->nl_net, table, fn);
1294 	/* Always release dst as dst->__refcnt is guaranteed
1295 	 * to be taken before entering this function
1296 	 */
1297 	dst_release_immediate(&rt->dst);
1298 	return err;
1299 }
1300 
1301 /*
1302  *	Routing tree lookup
1303  *
1304  */
1305 
1306 struct lookup_args {
1307 	int			offset;		/* key offset on rt6_info	*/
1308 	const struct in6_addr	*addr;		/* search key			*/
1309 };
1310 
1311 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1312 				       struct lookup_args *args)
1313 {
1314 	struct fib6_node *fn;
1315 	__be32 dir;
1316 
1317 	if (unlikely(args->offset == 0))
1318 		return NULL;
1319 
1320 	/*
1321 	 *	Descend on a tree
1322 	 */
1323 
1324 	fn = root;
1325 
1326 	for (;;) {
1327 		struct fib6_node *next;
1328 
1329 		dir = addr_bit_set(args->addr, fn->fn_bit);
1330 
1331 		next = dir ? rcu_dereference(fn->right) :
1332 			     rcu_dereference(fn->left);
1333 
1334 		if (next) {
1335 			fn = next;
1336 			continue;
1337 		}
1338 		break;
1339 	}
1340 
1341 	while (fn) {
1342 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1343 
1344 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1345 			struct rt6_info *leaf = rcu_dereference(fn->leaf);
1346 			struct rt6key *key;
1347 
1348 			if (!leaf)
1349 				goto backtrack;
1350 
1351 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1352 
1353 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1354 #ifdef CONFIG_IPV6_SUBTREES
1355 				if (subtree) {
1356 					struct fib6_node *sfn;
1357 					sfn = fib6_lookup_1(subtree, args + 1);
1358 					if (!sfn)
1359 						goto backtrack;
1360 					fn = sfn;
1361 				}
1362 #endif
1363 				if (fn->fn_flags & RTN_RTINFO)
1364 					return fn;
1365 			}
1366 		}
1367 backtrack:
1368 		if (fn->fn_flags & RTN_ROOT)
1369 			break;
1370 
1371 		fn = rcu_dereference(fn->parent);
1372 	}
1373 
1374 	return NULL;
1375 }
1376 
1377 /* called with rcu_read_lock() held
1378  */
1379 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1380 			      const struct in6_addr *saddr)
1381 {
1382 	struct fib6_node *fn;
1383 	struct lookup_args args[] = {
1384 		{
1385 			.offset = offsetof(struct rt6_info, rt6i_dst),
1386 			.addr = daddr,
1387 		},
1388 #ifdef CONFIG_IPV6_SUBTREES
1389 		{
1390 			.offset = offsetof(struct rt6_info, rt6i_src),
1391 			.addr = saddr,
1392 		},
1393 #endif
1394 		{
1395 			.offset = 0,	/* sentinel */
1396 		}
1397 	};
1398 
1399 	fn = fib6_lookup_1(root, daddr ? args : args + 1);
1400 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1401 		fn = root;
1402 
1403 	return fn;
1404 }
1405 
1406 /*
1407  *	Get node with specified destination prefix (and source prefix,
1408  *	if subtrees are used)
1409  *	exact_match == true means we try to find fn with exact match of
1410  *	the passed in prefix addr
1411  *	exact_match == false means we try to find fn with longest prefix
1412  *	match of the passed in prefix addr. This is useful for finding fn
1413  *	for cached route as it will be stored in the exception table under
1414  *	the node with longest prefix length.
1415  */
1416 
1417 
1418 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1419 				       const struct in6_addr *addr,
1420 				       int plen, int offset,
1421 				       bool exact_match)
1422 {
1423 	struct fib6_node *fn, *prev = NULL;
1424 
1425 	for (fn = root; fn ; ) {
1426 		struct rt6_info *leaf = rcu_dereference(fn->leaf);
1427 		struct rt6key *key;
1428 
1429 		/* This node is being deleted */
1430 		if (!leaf) {
1431 			if (plen <= fn->fn_bit)
1432 				goto out;
1433 			else
1434 				goto next;
1435 		}
1436 
1437 		key = (struct rt6key *)((u8 *)leaf + offset);
1438 
1439 		/*
1440 		 *	Prefix match
1441 		 */
1442 		if (plen < fn->fn_bit ||
1443 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1444 			goto out;
1445 
1446 		if (plen == fn->fn_bit)
1447 			return fn;
1448 
1449 		prev = fn;
1450 
1451 next:
1452 		/*
1453 		 *	We have more bits to go
1454 		 */
1455 		if (addr_bit_set(addr, fn->fn_bit))
1456 			fn = rcu_dereference(fn->right);
1457 		else
1458 			fn = rcu_dereference(fn->left);
1459 	}
1460 out:
1461 	if (exact_match)
1462 		return NULL;
1463 	else
1464 		return prev;
1465 }
1466 
1467 struct fib6_node *fib6_locate(struct fib6_node *root,
1468 			      const struct in6_addr *daddr, int dst_len,
1469 			      const struct in6_addr *saddr, int src_len,
1470 			      bool exact_match)
1471 {
1472 	struct fib6_node *fn;
1473 
1474 	fn = fib6_locate_1(root, daddr, dst_len,
1475 			   offsetof(struct rt6_info, rt6i_dst),
1476 			   exact_match);
1477 
1478 #ifdef CONFIG_IPV6_SUBTREES
1479 	if (src_len) {
1480 		WARN_ON(saddr == NULL);
1481 		if (fn) {
1482 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1483 
1484 			if (subtree) {
1485 				fn = fib6_locate_1(subtree, saddr, src_len,
1486 					   offsetof(struct rt6_info, rt6i_src),
1487 					   exact_match);
1488 			}
1489 		}
1490 	}
1491 #endif
1492 
1493 	if (fn && fn->fn_flags & RTN_RTINFO)
1494 		return fn;
1495 
1496 	return NULL;
1497 }
1498 
1499 
1500 /*
1501  *	Deletion
1502  *
1503  */
1504 
1505 static struct rt6_info *fib6_find_prefix(struct net *net,
1506 					 struct fib6_table *table,
1507 					 struct fib6_node *fn)
1508 {
1509 	struct fib6_node *child_left, *child_right;
1510 
1511 	if (fn->fn_flags & RTN_ROOT)
1512 		return net->ipv6.ip6_null_entry;
1513 
1514 	while (fn) {
1515 		child_left = rcu_dereference_protected(fn->left,
1516 				    lockdep_is_held(&table->tb6_lock));
1517 		child_right = rcu_dereference_protected(fn->right,
1518 				    lockdep_is_held(&table->tb6_lock));
1519 		if (child_left)
1520 			return rcu_dereference_protected(child_left->leaf,
1521 					lockdep_is_held(&table->tb6_lock));
1522 		if (child_right)
1523 			return rcu_dereference_protected(child_right->leaf,
1524 					lockdep_is_held(&table->tb6_lock));
1525 
1526 		fn = FIB6_SUBTREE(fn);
1527 	}
1528 	return NULL;
1529 }
1530 
1531 /*
1532  *	Called to trim the tree of intermediate nodes when possible. "fn"
1533  *	is the node we want to try and remove.
1534  *	Need to own table->tb6_lock
1535  */
1536 
1537 static struct fib6_node *fib6_repair_tree(struct net *net,
1538 					  struct fib6_table *table,
1539 					  struct fib6_node *fn)
1540 {
1541 	int children;
1542 	int nstate;
1543 	struct fib6_node *child;
1544 	struct fib6_walker *w;
1545 	int iter = 0;
1546 
1547 	/* Set fn->leaf to null_entry for root node. */
1548 	if (fn->fn_flags & RTN_TL_ROOT) {
1549 		rcu_assign_pointer(fn->leaf, net->ipv6.ip6_null_entry);
1550 		return fn;
1551 	}
1552 
1553 	for (;;) {
1554 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1555 					    lockdep_is_held(&table->tb6_lock));
1556 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1557 					    lockdep_is_held(&table->tb6_lock));
1558 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1559 					    lockdep_is_held(&table->tb6_lock));
1560 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1561 					    lockdep_is_held(&table->tb6_lock));
1562 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1563 					    lockdep_is_held(&table->tb6_lock));
1564 		struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1565 					    lockdep_is_held(&table->tb6_lock));
1566 		struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1567 					    lockdep_is_held(&table->tb6_lock));
1568 		struct rt6_info *new_fn_leaf;
1569 
1570 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1571 		iter++;
1572 
1573 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1574 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1575 		WARN_ON(fn_leaf);
1576 
1577 		children = 0;
1578 		child = NULL;
1579 		if (fn_r)
1580 			child = fn_r, children |= 1;
1581 		if (fn_l)
1582 			child = fn_l, children |= 2;
1583 
1584 		if (children == 3 || FIB6_SUBTREE(fn)
1585 #ifdef CONFIG_IPV6_SUBTREES
1586 		    /* Subtree root (i.e. fn) may have one child */
1587 		    || (children && fn->fn_flags & RTN_ROOT)
1588 #endif
1589 		    ) {
1590 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1591 #if RT6_DEBUG >= 2
1592 			if (!new_fn_leaf) {
1593 				WARN_ON(!new_fn_leaf);
1594 				new_fn_leaf = net->ipv6.ip6_null_entry;
1595 			}
1596 #endif
1597 			atomic_inc(&new_fn_leaf->rt6i_ref);
1598 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1599 			return pn;
1600 		}
1601 
1602 #ifdef CONFIG_IPV6_SUBTREES
1603 		if (FIB6_SUBTREE(pn) == fn) {
1604 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1605 			RCU_INIT_POINTER(pn->subtree, NULL);
1606 			nstate = FWS_L;
1607 		} else {
1608 			WARN_ON(fn->fn_flags & RTN_ROOT);
1609 #endif
1610 			if (pn_r == fn)
1611 				rcu_assign_pointer(pn->right, child);
1612 			else if (pn_l == fn)
1613 				rcu_assign_pointer(pn->left, child);
1614 #if RT6_DEBUG >= 2
1615 			else
1616 				WARN_ON(1);
1617 #endif
1618 			if (child)
1619 				rcu_assign_pointer(child->parent, pn);
1620 			nstate = FWS_R;
1621 #ifdef CONFIG_IPV6_SUBTREES
1622 		}
1623 #endif
1624 
1625 		read_lock(&net->ipv6.fib6_walker_lock);
1626 		FOR_WALKERS(net, w) {
1627 			if (!child) {
1628 				if (w->node == fn) {
1629 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1630 					w->node = pn;
1631 					w->state = nstate;
1632 				}
1633 			} else {
1634 				if (w->node == fn) {
1635 					w->node = child;
1636 					if (children&2) {
1637 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1638 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1639 					} else {
1640 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1641 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1642 					}
1643 				}
1644 			}
1645 		}
1646 		read_unlock(&net->ipv6.fib6_walker_lock);
1647 
1648 		node_free(net, fn);
1649 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1650 			return pn;
1651 
1652 		RCU_INIT_POINTER(pn->leaf, NULL);
1653 		rt6_release(pn_leaf);
1654 		fn = pn;
1655 	}
1656 }
1657 
1658 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1659 			   struct rt6_info __rcu **rtp, struct nl_info *info)
1660 {
1661 	struct fib6_walker *w;
1662 	struct rt6_info *rt = rcu_dereference_protected(*rtp,
1663 				    lockdep_is_held(&table->tb6_lock));
1664 	struct net *net = info->nl_net;
1665 
1666 	RT6_TRACE("fib6_del_route\n");
1667 
1668 	WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1669 
1670 	/* Unlink it */
1671 	*rtp = rt->rt6_next;
1672 	rt->rt6i_node = NULL;
1673 	net->ipv6.rt6_stats->fib_rt_entries--;
1674 	net->ipv6.rt6_stats->fib_discarded_routes++;
1675 
1676 	/* Flush all cached dst in exception table */
1677 	rt6_flush_exceptions(rt);
1678 
1679 	/* Reset round-robin state, if necessary */
1680 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1681 		fn->rr_ptr = NULL;
1682 
1683 	/* Remove this entry from other siblings */
1684 	if (rt->rt6i_nsiblings) {
1685 		struct rt6_info *sibling, *next_sibling;
1686 
1687 		list_for_each_entry_safe(sibling, next_sibling,
1688 					 &rt->rt6i_siblings, rt6i_siblings)
1689 			sibling->rt6i_nsiblings--;
1690 		rt->rt6i_nsiblings = 0;
1691 		list_del_init(&rt->rt6i_siblings);
1692 		rt6_multipath_rebalance(next_sibling);
1693 	}
1694 
1695 	/* Adjust walkers */
1696 	read_lock(&net->ipv6.fib6_walker_lock);
1697 	FOR_WALKERS(net, w) {
1698 		if (w->state == FWS_C && w->leaf == rt) {
1699 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1700 			w->leaf = rcu_dereference_protected(rt->rt6_next,
1701 					    lockdep_is_held(&table->tb6_lock));
1702 			if (!w->leaf)
1703 				w->state = FWS_U;
1704 		}
1705 	}
1706 	read_unlock(&net->ipv6.fib6_walker_lock);
1707 
1708 	/* If it was last route, call fib6_repair_tree() to:
1709 	 * 1. For root node, put back null_entry as how the table was created.
1710 	 * 2. For other nodes, expunge its radix tree node.
1711 	 */
1712 	if (!rcu_access_pointer(fn->leaf)) {
1713 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1714 			fn->fn_flags &= ~RTN_RTINFO;
1715 			net->ipv6.rt6_stats->fib_route_nodes--;
1716 		}
1717 		fn = fib6_repair_tree(net, table, fn);
1718 	}
1719 
1720 	fib6_purge_rt(rt, fn, net);
1721 
1722 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1723 	if (!info->skip_notify)
1724 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1725 	rt6_release(rt);
1726 }
1727 
1728 /* Need to own table->tb6_lock */
1729 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1730 {
1731 	struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1732 				    lockdep_is_held(&rt->rt6i_table->tb6_lock));
1733 	struct fib6_table *table = rt->rt6i_table;
1734 	struct net *net = info->nl_net;
1735 	struct rt6_info __rcu **rtp;
1736 	struct rt6_info __rcu **rtp_next;
1737 
1738 #if RT6_DEBUG >= 2
1739 	if (rt->dst.obsolete > 0) {
1740 		WARN_ON(fn);
1741 		return -ENOENT;
1742 	}
1743 #endif
1744 	if (!fn || rt == net->ipv6.ip6_null_entry)
1745 		return -ENOENT;
1746 
1747 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1748 
1749 	/* remove cached dst from exception table */
1750 	if (rt->rt6i_flags & RTF_CACHE)
1751 		return rt6_remove_exception_rt(rt);
1752 
1753 	/*
1754 	 *	Walk the leaf entries looking for ourself
1755 	 */
1756 
1757 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1758 		struct rt6_info *cur = rcu_dereference_protected(*rtp,
1759 					lockdep_is_held(&table->tb6_lock));
1760 		if (rt == cur) {
1761 			fib6_del_route(table, fn, rtp, info);
1762 			return 0;
1763 		}
1764 		rtp_next = &cur->rt6_next;
1765 	}
1766 	return -ENOENT;
1767 }
1768 
1769 /*
1770  *	Tree traversal function.
1771  *
1772  *	Certainly, it is not interrupt safe.
1773  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1774  *	It means, that we can modify tree during walking
1775  *	and use this function for garbage collection, clone pruning,
1776  *	cleaning tree when a device goes down etc. etc.
1777  *
1778  *	It guarantees that every node will be traversed,
1779  *	and that it will be traversed only once.
1780  *
1781  *	Callback function w->func may return:
1782  *	0 -> continue walking.
1783  *	positive value -> walking is suspended (used by tree dumps,
1784  *	and probably by gc, if it will be split to several slices)
1785  *	negative value -> terminate walking.
1786  *
1787  *	The function itself returns:
1788  *	0   -> walk is complete.
1789  *	>0  -> walk is incomplete (i.e. suspended)
1790  *	<0  -> walk is terminated by an error.
1791  *
1792  *	This function is called with tb6_lock held.
1793  */
1794 
1795 static int fib6_walk_continue(struct fib6_walker *w)
1796 {
1797 	struct fib6_node *fn, *pn, *left, *right;
1798 
1799 	/* w->root should always be table->tb6_root */
1800 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1801 
1802 	for (;;) {
1803 		fn = w->node;
1804 		if (!fn)
1805 			return 0;
1806 
1807 		switch (w->state) {
1808 #ifdef CONFIG_IPV6_SUBTREES
1809 		case FWS_S:
1810 			if (FIB6_SUBTREE(fn)) {
1811 				w->node = FIB6_SUBTREE(fn);
1812 				continue;
1813 			}
1814 			w->state = FWS_L;
1815 #endif
1816 			/* fall through */
1817 		case FWS_L:
1818 			left = rcu_dereference_protected(fn->left, 1);
1819 			if (left) {
1820 				w->node = left;
1821 				w->state = FWS_INIT;
1822 				continue;
1823 			}
1824 			w->state = FWS_R;
1825 			/* fall through */
1826 		case FWS_R:
1827 			right = rcu_dereference_protected(fn->right, 1);
1828 			if (right) {
1829 				w->node = right;
1830 				w->state = FWS_INIT;
1831 				continue;
1832 			}
1833 			w->state = FWS_C;
1834 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1835 			/* fall through */
1836 		case FWS_C:
1837 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1838 				int err;
1839 
1840 				if (w->skip) {
1841 					w->skip--;
1842 					goto skip;
1843 				}
1844 
1845 				err = w->func(w);
1846 				if (err)
1847 					return err;
1848 
1849 				w->count++;
1850 				continue;
1851 			}
1852 skip:
1853 			w->state = FWS_U;
1854 			/* fall through */
1855 		case FWS_U:
1856 			if (fn == w->root)
1857 				return 0;
1858 			pn = rcu_dereference_protected(fn->parent, 1);
1859 			left = rcu_dereference_protected(pn->left, 1);
1860 			right = rcu_dereference_protected(pn->right, 1);
1861 			w->node = pn;
1862 #ifdef CONFIG_IPV6_SUBTREES
1863 			if (FIB6_SUBTREE(pn) == fn) {
1864 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1865 				w->state = FWS_L;
1866 				continue;
1867 			}
1868 #endif
1869 			if (left == fn) {
1870 				w->state = FWS_R;
1871 				continue;
1872 			}
1873 			if (right == fn) {
1874 				w->state = FWS_C;
1875 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1876 				continue;
1877 			}
1878 #if RT6_DEBUG >= 2
1879 			WARN_ON(1);
1880 #endif
1881 		}
1882 	}
1883 }
1884 
1885 static int fib6_walk(struct net *net, struct fib6_walker *w)
1886 {
1887 	int res;
1888 
1889 	w->state = FWS_INIT;
1890 	w->node = w->root;
1891 
1892 	fib6_walker_link(net, w);
1893 	res = fib6_walk_continue(w);
1894 	if (res <= 0)
1895 		fib6_walker_unlink(net, w);
1896 	return res;
1897 }
1898 
1899 static int fib6_clean_node(struct fib6_walker *w)
1900 {
1901 	int res;
1902 	struct rt6_info *rt;
1903 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1904 	struct nl_info info = {
1905 		.nl_net = c->net,
1906 	};
1907 
1908 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1909 	    w->node->fn_sernum != c->sernum)
1910 		w->node->fn_sernum = c->sernum;
1911 
1912 	if (!c->func) {
1913 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1914 		w->leaf = NULL;
1915 		return 0;
1916 	}
1917 
1918 	for_each_fib6_walker_rt(w) {
1919 		res = c->func(rt, c->arg);
1920 		if (res == -1) {
1921 			w->leaf = rt;
1922 			res = fib6_del(rt, &info);
1923 			if (res) {
1924 #if RT6_DEBUG >= 2
1925 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1926 					 __func__, rt,
1927 					 rcu_access_pointer(rt->rt6i_node),
1928 					 res);
1929 #endif
1930 				continue;
1931 			}
1932 			return 0;
1933 		} else if (res == -2) {
1934 			if (WARN_ON(!rt->rt6i_nsiblings))
1935 				continue;
1936 			rt = list_last_entry(&rt->rt6i_siblings,
1937 					     struct rt6_info, rt6i_siblings);
1938 			continue;
1939 		}
1940 		WARN_ON(res != 0);
1941 	}
1942 	w->leaf = rt;
1943 	return 0;
1944 }
1945 
1946 /*
1947  *	Convenient frontend to tree walker.
1948  *
1949  *	func is called on each route.
1950  *		It may return -2 -> skip multipath route.
1951  *			      -1 -> delete this route.
1952  *		              0  -> continue walking
1953  */
1954 
1955 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1956 			    int (*func)(struct rt6_info *, void *arg),
1957 			    int sernum, void *arg)
1958 {
1959 	struct fib6_cleaner c;
1960 
1961 	c.w.root = root;
1962 	c.w.func = fib6_clean_node;
1963 	c.w.count = 0;
1964 	c.w.skip = 0;
1965 	c.func = func;
1966 	c.sernum = sernum;
1967 	c.arg = arg;
1968 	c.net = net;
1969 
1970 	fib6_walk(net, &c.w);
1971 }
1972 
1973 static void __fib6_clean_all(struct net *net,
1974 			     int (*func)(struct rt6_info *, void *),
1975 			     int sernum, void *arg)
1976 {
1977 	struct fib6_table *table;
1978 	struct hlist_head *head;
1979 	unsigned int h;
1980 
1981 	rcu_read_lock();
1982 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1983 		head = &net->ipv6.fib_table_hash[h];
1984 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1985 			spin_lock_bh(&table->tb6_lock);
1986 			fib6_clean_tree(net, &table->tb6_root,
1987 					func, sernum, arg);
1988 			spin_unlock_bh(&table->tb6_lock);
1989 		}
1990 	}
1991 	rcu_read_unlock();
1992 }
1993 
1994 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1995 		    void *arg)
1996 {
1997 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1998 }
1999 
2000 static void fib6_flush_trees(struct net *net)
2001 {
2002 	int new_sernum = fib6_new_sernum(net);
2003 
2004 	__fib6_clean_all(net, NULL, new_sernum, NULL);
2005 }
2006 
2007 /*
2008  *	Garbage collection
2009  */
2010 
2011 static int fib6_age(struct rt6_info *rt, void *arg)
2012 {
2013 	struct fib6_gc_args *gc_args = arg;
2014 	unsigned long now = jiffies;
2015 
2016 	/*
2017 	 *	check addrconf expiration here.
2018 	 *	Routes are expired even if they are in use.
2019 	 */
2020 
2021 	if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
2022 		if (time_after(now, rt->dst.expires)) {
2023 			RT6_TRACE("expiring %p\n", rt);
2024 			return -1;
2025 		}
2026 		gc_args->more++;
2027 	}
2028 
2029 	/*	Also age clones in the exception table.
2030 	 *	Note, that clones are aged out
2031 	 *	only if they are not in use now.
2032 	 */
2033 	rt6_age_exceptions(rt, gc_args, now);
2034 
2035 	return 0;
2036 }
2037 
2038 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2039 {
2040 	struct fib6_gc_args gc_args;
2041 	unsigned long now;
2042 
2043 	if (force) {
2044 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2045 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2046 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2047 		return;
2048 	}
2049 	gc_args.timeout = expires ? (int)expires :
2050 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2051 	gc_args.more = 0;
2052 
2053 	fib6_clean_all(net, fib6_age, &gc_args);
2054 	now = jiffies;
2055 	net->ipv6.ip6_rt_last_gc = now;
2056 
2057 	if (gc_args.more)
2058 		mod_timer(&net->ipv6.ip6_fib_timer,
2059 			  round_jiffies(now
2060 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2061 	else
2062 		del_timer(&net->ipv6.ip6_fib_timer);
2063 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2064 }
2065 
2066 static void fib6_gc_timer_cb(struct timer_list *t)
2067 {
2068 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2069 
2070 	fib6_run_gc(0, arg, true);
2071 }
2072 
2073 static int __net_init fib6_net_init(struct net *net)
2074 {
2075 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2076 	int err;
2077 
2078 	err = fib6_notifier_init(net);
2079 	if (err)
2080 		return err;
2081 
2082 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2083 	rwlock_init(&net->ipv6.fib6_walker_lock);
2084 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2085 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2086 
2087 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2088 	if (!net->ipv6.rt6_stats)
2089 		goto out_timer;
2090 
2091 	/* Avoid false sharing : Use at least a full cache line */
2092 	size = max_t(size_t, size, L1_CACHE_BYTES);
2093 
2094 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2095 	if (!net->ipv6.fib_table_hash)
2096 		goto out_rt6_stats;
2097 
2098 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2099 					  GFP_KERNEL);
2100 	if (!net->ipv6.fib6_main_tbl)
2101 		goto out_fib_table_hash;
2102 
2103 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2104 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2105 			   net->ipv6.ip6_null_entry);
2106 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2107 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2108 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2109 
2110 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2111 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2112 					   GFP_KERNEL);
2113 	if (!net->ipv6.fib6_local_tbl)
2114 		goto out_fib6_main_tbl;
2115 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2116 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2117 			   net->ipv6.ip6_null_entry);
2118 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2119 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2120 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2121 #endif
2122 	fib6_tables_init(net);
2123 
2124 	return 0;
2125 
2126 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2127 out_fib6_main_tbl:
2128 	kfree(net->ipv6.fib6_main_tbl);
2129 #endif
2130 out_fib_table_hash:
2131 	kfree(net->ipv6.fib_table_hash);
2132 out_rt6_stats:
2133 	kfree(net->ipv6.rt6_stats);
2134 out_timer:
2135 	fib6_notifier_exit(net);
2136 	return -ENOMEM;
2137 }
2138 
2139 static void fib6_net_exit(struct net *net)
2140 {
2141 	unsigned int i;
2142 
2143 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2144 
2145 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2146 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2147 		struct hlist_node *tmp;
2148 		struct fib6_table *tb;
2149 
2150 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2151 			hlist_del(&tb->tb6_hlist);
2152 			fib6_free_table(tb);
2153 		}
2154 	}
2155 
2156 	kfree(net->ipv6.fib_table_hash);
2157 	kfree(net->ipv6.rt6_stats);
2158 	fib6_notifier_exit(net);
2159 }
2160 
2161 static struct pernet_operations fib6_net_ops = {
2162 	.init = fib6_net_init,
2163 	.exit = fib6_net_exit,
2164 	.async = true,
2165 };
2166 
2167 int __init fib6_init(void)
2168 {
2169 	int ret = -ENOMEM;
2170 
2171 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2172 					   sizeof(struct fib6_node),
2173 					   0, SLAB_HWCACHE_ALIGN,
2174 					   NULL);
2175 	if (!fib6_node_kmem)
2176 		goto out;
2177 
2178 	ret = register_pernet_subsys(&fib6_net_ops);
2179 	if (ret)
2180 		goto out_kmem_cache_create;
2181 
2182 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2183 				   inet6_dump_fib, 0);
2184 	if (ret)
2185 		goto out_unregister_subsys;
2186 
2187 	__fib6_flush_trees = fib6_flush_trees;
2188 out:
2189 	return ret;
2190 
2191 out_unregister_subsys:
2192 	unregister_pernet_subsys(&fib6_net_ops);
2193 out_kmem_cache_create:
2194 	kmem_cache_destroy(fib6_node_kmem);
2195 	goto out;
2196 }
2197 
2198 void fib6_gc_cleanup(void)
2199 {
2200 	unregister_pernet_subsys(&fib6_net_ops);
2201 	kmem_cache_destroy(fib6_node_kmem);
2202 }
2203 
2204 #ifdef CONFIG_PROC_FS
2205 
2206 struct ipv6_route_iter {
2207 	struct seq_net_private p;
2208 	struct fib6_walker w;
2209 	loff_t skip;
2210 	struct fib6_table *tbl;
2211 	int sernum;
2212 };
2213 
2214 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2215 {
2216 	struct rt6_info *rt = v;
2217 	struct ipv6_route_iter *iter = seq->private;
2218 
2219 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2220 
2221 #ifdef CONFIG_IPV6_SUBTREES
2222 	seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2223 #else
2224 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2225 #endif
2226 	if (rt->rt6i_flags & RTF_GATEWAY)
2227 		seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2228 	else
2229 		seq_puts(seq, "00000000000000000000000000000000");
2230 
2231 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2232 		   rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2233 		   rt->dst.__use, rt->rt6i_flags,
2234 		   rt->dst.dev ? rt->dst.dev->name : "");
2235 	iter->w.leaf = NULL;
2236 	return 0;
2237 }
2238 
2239 static int ipv6_route_yield(struct fib6_walker *w)
2240 {
2241 	struct ipv6_route_iter *iter = w->args;
2242 
2243 	if (!iter->skip)
2244 		return 1;
2245 
2246 	do {
2247 		iter->w.leaf = rcu_dereference_protected(
2248 				iter->w.leaf->rt6_next,
2249 				lockdep_is_held(&iter->tbl->tb6_lock));
2250 		iter->skip--;
2251 		if (!iter->skip && iter->w.leaf)
2252 			return 1;
2253 	} while (iter->w.leaf);
2254 
2255 	return 0;
2256 }
2257 
2258 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2259 				      struct net *net)
2260 {
2261 	memset(&iter->w, 0, sizeof(iter->w));
2262 	iter->w.func = ipv6_route_yield;
2263 	iter->w.root = &iter->tbl->tb6_root;
2264 	iter->w.state = FWS_INIT;
2265 	iter->w.node = iter->w.root;
2266 	iter->w.args = iter;
2267 	iter->sernum = iter->w.root->fn_sernum;
2268 	INIT_LIST_HEAD(&iter->w.lh);
2269 	fib6_walker_link(net, &iter->w);
2270 }
2271 
2272 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2273 						    struct net *net)
2274 {
2275 	unsigned int h;
2276 	struct hlist_node *node;
2277 
2278 	if (tbl) {
2279 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2280 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2281 	} else {
2282 		h = 0;
2283 		node = NULL;
2284 	}
2285 
2286 	while (!node && h < FIB6_TABLE_HASHSZ) {
2287 		node = rcu_dereference_bh(
2288 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2289 	}
2290 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2291 }
2292 
2293 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2294 {
2295 	if (iter->sernum != iter->w.root->fn_sernum) {
2296 		iter->sernum = iter->w.root->fn_sernum;
2297 		iter->w.state = FWS_INIT;
2298 		iter->w.node = iter->w.root;
2299 		WARN_ON(iter->w.skip);
2300 		iter->w.skip = iter->w.count;
2301 	}
2302 }
2303 
2304 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2305 {
2306 	int r;
2307 	struct rt6_info *n;
2308 	struct net *net = seq_file_net(seq);
2309 	struct ipv6_route_iter *iter = seq->private;
2310 
2311 	if (!v)
2312 		goto iter_table;
2313 
2314 	n = rcu_dereference_bh(((struct rt6_info *)v)->rt6_next);
2315 	if (n) {
2316 		++*pos;
2317 		return n;
2318 	}
2319 
2320 iter_table:
2321 	ipv6_route_check_sernum(iter);
2322 	spin_lock_bh(&iter->tbl->tb6_lock);
2323 	r = fib6_walk_continue(&iter->w);
2324 	spin_unlock_bh(&iter->tbl->tb6_lock);
2325 	if (r > 0) {
2326 		if (v)
2327 			++*pos;
2328 		return iter->w.leaf;
2329 	} else if (r < 0) {
2330 		fib6_walker_unlink(net, &iter->w);
2331 		return NULL;
2332 	}
2333 	fib6_walker_unlink(net, &iter->w);
2334 
2335 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2336 	if (!iter->tbl)
2337 		return NULL;
2338 
2339 	ipv6_route_seq_setup_walk(iter, net);
2340 	goto iter_table;
2341 }
2342 
2343 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2344 	__acquires(RCU_BH)
2345 {
2346 	struct net *net = seq_file_net(seq);
2347 	struct ipv6_route_iter *iter = seq->private;
2348 
2349 	rcu_read_lock_bh();
2350 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2351 	iter->skip = *pos;
2352 
2353 	if (iter->tbl) {
2354 		ipv6_route_seq_setup_walk(iter, net);
2355 		return ipv6_route_seq_next(seq, NULL, pos);
2356 	} else {
2357 		return NULL;
2358 	}
2359 }
2360 
2361 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2362 {
2363 	struct fib6_walker *w = &iter->w;
2364 	return w->node && !(w->state == FWS_U && w->node == w->root);
2365 }
2366 
2367 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2368 	__releases(RCU_BH)
2369 {
2370 	struct net *net = seq_file_net(seq);
2371 	struct ipv6_route_iter *iter = seq->private;
2372 
2373 	if (ipv6_route_iter_active(iter))
2374 		fib6_walker_unlink(net, &iter->w);
2375 
2376 	rcu_read_unlock_bh();
2377 }
2378 
2379 static const struct seq_operations ipv6_route_seq_ops = {
2380 	.start	= ipv6_route_seq_start,
2381 	.next	= ipv6_route_seq_next,
2382 	.stop	= ipv6_route_seq_stop,
2383 	.show	= ipv6_route_seq_show
2384 };
2385 
2386 int ipv6_route_open(struct inode *inode, struct file *file)
2387 {
2388 	return seq_open_net(inode, file, &ipv6_route_seq_ops,
2389 			    sizeof(struct ipv6_route_iter));
2390 }
2391 
2392 #endif /* CONFIG_PROC_FS */
2393