xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision bb3982b4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Linux INET6 implementation
4  *	Forwarding Information Database
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Changes:
10  *	Yuji SEKIYA @USAGI:	Support default route on router node;
11  *				remove ip6_null_entry from the top of
12  *				routing table.
13  *	Ville Nuorvala:		Fixed routing subtrees.
14  */
15 
16 #define pr_fmt(fmt) "IPv6: " fmt
17 
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/net.h>
21 #include <linux/route.h>
22 #include <linux/netdevice.h>
23 #include <linux/in6.h>
24 #include <linux/init.h>
25 #include <linux/list.h>
26 #include <linux/slab.h>
27 
28 #include <net/ip.h>
29 #include <net/ipv6.h>
30 #include <net/ndisc.h>
31 #include <net/addrconf.h>
32 #include <net/lwtunnel.h>
33 #include <net/fib_notifier.h>
34 
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
37 
38 static struct kmem_cache *fib6_node_kmem __read_mostly;
39 
40 struct fib6_cleaner {
41 	struct fib6_walker w;
42 	struct net *net;
43 	int (*func)(struct fib6_info *, void *arg);
44 	int sernum;
45 	void *arg;
46 	bool skip_notify;
47 };
48 
49 #ifdef CONFIG_IPV6_SUBTREES
50 #define FWS_INIT FWS_S
51 #else
52 #define FWS_INIT FWS_L
53 #endif
54 
55 static struct fib6_info *fib6_find_prefix(struct net *net,
56 					 struct fib6_table *table,
57 					 struct fib6_node *fn);
58 static struct fib6_node *fib6_repair_tree(struct net *net,
59 					  struct fib6_table *table,
60 					  struct fib6_node *fn);
61 static int fib6_walk(struct net *net, struct fib6_walker *w);
62 static int fib6_walk_continue(struct fib6_walker *w);
63 
64 /*
65  *	A routing update causes an increase of the serial number on the
66  *	affected subtree. This allows for cached routes to be asynchronously
67  *	tested when modifications are made to the destination cache as a
68  *	result of redirects, path MTU changes, etc.
69  */
70 
71 static void fib6_gc_timer_cb(struct timer_list *t);
72 
73 #define FOR_WALKERS(net, w) \
74 	list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
75 
76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
77 {
78 	write_lock_bh(&net->ipv6.fib6_walker_lock);
79 	list_add(&w->lh, &net->ipv6.fib6_walkers);
80 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
81 }
82 
83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
84 {
85 	write_lock_bh(&net->ipv6.fib6_walker_lock);
86 	list_del(&w->lh);
87 	write_unlock_bh(&net->ipv6.fib6_walker_lock);
88 }
89 
90 static int fib6_new_sernum(struct net *net)
91 {
92 	int new, old;
93 
94 	do {
95 		old = atomic_read(&net->ipv6.fib6_sernum);
96 		new = old < INT_MAX ? old + 1 : 1;
97 	} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
98 				old, new) != old);
99 	return new;
100 }
101 
102 enum {
103 	FIB6_NO_SERNUM_CHANGE = 0,
104 };
105 
106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
107 {
108 	struct fib6_node *fn;
109 
110 	fn = rcu_dereference_protected(f6i->fib6_node,
111 			lockdep_is_held(&f6i->fib6_table->tb6_lock));
112 	if (fn)
113 		fn->fn_sernum = fib6_new_sernum(net);
114 }
115 
116 /*
117  *	Auxiliary address test functions for the radix tree.
118  *
119  *	These assume a 32bit processor (although it will work on
120  *	64bit processors)
121  */
122 
123 /*
124  *	test bit
125  */
126 #if defined(__LITTLE_ENDIAN)
127 # define BITOP_BE32_SWIZZLE	(0x1F & ~7)
128 #else
129 # define BITOP_BE32_SWIZZLE	0
130 #endif
131 
132 static __be32 addr_bit_set(const void *token, int fn_bit)
133 {
134 	const __be32 *addr = token;
135 	/*
136 	 * Here,
137 	 *	1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
138 	 * is optimized version of
139 	 *	htonl(1 << ((~fn_bit)&0x1F))
140 	 * See include/asm-generic/bitops/le.h.
141 	 */
142 	return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
143 	       addr[fn_bit >> 5];
144 }
145 
146 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
147 {
148 	struct fib6_info *f6i;
149 
150 	f6i = kzalloc(sizeof(*f6i), gfp_flags);
151 	if (!f6i)
152 		return NULL;
153 
154 	f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
155 	if (!f6i->rt6i_pcpu) {
156 		kfree(f6i);
157 		return NULL;
158 	}
159 
160 	INIT_LIST_HEAD(&f6i->fib6_siblings);
161 	refcount_set(&f6i->fib6_ref, 1);
162 
163 	return f6i;
164 }
165 
166 void fib6_info_destroy_rcu(struct rcu_head *head)
167 {
168 	struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
169 	struct rt6_exception_bucket *bucket;
170 
171 	WARN_ON(f6i->fib6_node);
172 
173 	bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
174 	kfree(bucket);
175 
176 	if (f6i->rt6i_pcpu) {
177 		int cpu;
178 
179 		for_each_possible_cpu(cpu) {
180 			struct rt6_info **ppcpu_rt;
181 			struct rt6_info *pcpu_rt;
182 
183 			ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
184 			pcpu_rt = *ppcpu_rt;
185 			if (pcpu_rt) {
186 				dst_dev_put(&pcpu_rt->dst);
187 				dst_release(&pcpu_rt->dst);
188 				*ppcpu_rt = NULL;
189 			}
190 		}
191 
192 		free_percpu(f6i->rt6i_pcpu);
193 	}
194 
195 	fib6_nh_release(&f6i->fib6_nh);
196 
197 	ip_fib_metrics_put(f6i->fib6_metrics);
198 
199 	kfree(f6i);
200 }
201 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
202 
203 static struct fib6_node *node_alloc(struct net *net)
204 {
205 	struct fib6_node *fn;
206 
207 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
208 	if (fn)
209 		net->ipv6.rt6_stats->fib_nodes++;
210 
211 	return fn;
212 }
213 
214 static void node_free_immediate(struct net *net, struct fib6_node *fn)
215 {
216 	kmem_cache_free(fib6_node_kmem, fn);
217 	net->ipv6.rt6_stats->fib_nodes--;
218 }
219 
220 static void node_free_rcu(struct rcu_head *head)
221 {
222 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
223 
224 	kmem_cache_free(fib6_node_kmem, fn);
225 }
226 
227 static void node_free(struct net *net, struct fib6_node *fn)
228 {
229 	call_rcu(&fn->rcu, node_free_rcu);
230 	net->ipv6.rt6_stats->fib_nodes--;
231 }
232 
233 static void fib6_free_table(struct fib6_table *table)
234 {
235 	inetpeer_invalidate_tree(&table->tb6_peers);
236 	kfree(table);
237 }
238 
239 static void fib6_link_table(struct net *net, struct fib6_table *tb)
240 {
241 	unsigned int h;
242 
243 	/*
244 	 * Initialize table lock at a single place to give lockdep a key,
245 	 * tables aren't visible prior to being linked to the list.
246 	 */
247 	spin_lock_init(&tb->tb6_lock);
248 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
249 
250 	/*
251 	 * No protection necessary, this is the only list mutatation
252 	 * operation, tables never disappear once they exist.
253 	 */
254 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
255 }
256 
257 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
258 
259 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
260 {
261 	struct fib6_table *table;
262 
263 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
264 	if (table) {
265 		table->tb6_id = id;
266 		rcu_assign_pointer(table->tb6_root.leaf,
267 				   net->ipv6.fib6_null_entry);
268 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
269 		inet_peer_base_init(&table->tb6_peers);
270 	}
271 
272 	return table;
273 }
274 
275 struct fib6_table *fib6_new_table(struct net *net, u32 id)
276 {
277 	struct fib6_table *tb;
278 
279 	if (id == 0)
280 		id = RT6_TABLE_MAIN;
281 	tb = fib6_get_table(net, id);
282 	if (tb)
283 		return tb;
284 
285 	tb = fib6_alloc_table(net, id);
286 	if (tb)
287 		fib6_link_table(net, tb);
288 
289 	return tb;
290 }
291 EXPORT_SYMBOL_GPL(fib6_new_table);
292 
293 struct fib6_table *fib6_get_table(struct net *net, u32 id)
294 {
295 	struct fib6_table *tb;
296 	struct hlist_head *head;
297 	unsigned int h;
298 
299 	if (id == 0)
300 		id = RT6_TABLE_MAIN;
301 	h = id & (FIB6_TABLE_HASHSZ - 1);
302 	rcu_read_lock();
303 	head = &net->ipv6.fib_table_hash[h];
304 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
305 		if (tb->tb6_id == id) {
306 			rcu_read_unlock();
307 			return tb;
308 		}
309 	}
310 	rcu_read_unlock();
311 
312 	return NULL;
313 }
314 EXPORT_SYMBOL_GPL(fib6_get_table);
315 
316 static void __net_init fib6_tables_init(struct net *net)
317 {
318 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
319 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
320 }
321 #else
322 
323 struct fib6_table *fib6_new_table(struct net *net, u32 id)
324 {
325 	return fib6_get_table(net, id);
326 }
327 
328 struct fib6_table *fib6_get_table(struct net *net, u32 id)
329 {
330 	  return net->ipv6.fib6_main_tbl;
331 }
332 
333 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
334 				   const struct sk_buff *skb,
335 				   int flags, pol_lookup_t lookup)
336 {
337 	struct rt6_info *rt;
338 
339 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
340 	if (rt->dst.error == -EAGAIN) {
341 		ip6_rt_put(rt);
342 		rt = net->ipv6.ip6_null_entry;
343 		dst_hold(&rt->dst);
344 	}
345 
346 	return &rt->dst;
347 }
348 
349 /* called with rcu lock held; no reference taken on fib6_info */
350 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
351 		struct fib6_result *res, int flags)
352 {
353 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
354 				 res, flags);
355 }
356 
357 static void __net_init fib6_tables_init(struct net *net)
358 {
359 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
360 }
361 
362 #endif
363 
364 unsigned int fib6_tables_seq_read(struct net *net)
365 {
366 	unsigned int h, fib_seq = 0;
367 
368 	rcu_read_lock();
369 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
370 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
371 		struct fib6_table *tb;
372 
373 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
374 			fib_seq += tb->fib_seq;
375 	}
376 	rcu_read_unlock();
377 
378 	return fib_seq;
379 }
380 
381 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
382 				    enum fib_event_type event_type,
383 				    struct fib6_info *rt)
384 {
385 	struct fib6_entry_notifier_info info = {
386 		.rt = rt,
387 	};
388 
389 	return call_fib6_notifier(nb, net, event_type, &info.info);
390 }
391 
392 static int call_fib6_entry_notifiers(struct net *net,
393 				     enum fib_event_type event_type,
394 				     struct fib6_info *rt,
395 				     struct netlink_ext_ack *extack)
396 {
397 	struct fib6_entry_notifier_info info = {
398 		.info.extack = extack,
399 		.rt = rt,
400 	};
401 
402 	rt->fib6_table->fib_seq++;
403 	return call_fib6_notifiers(net, event_type, &info.info);
404 }
405 
406 struct fib6_dump_arg {
407 	struct net *net;
408 	struct notifier_block *nb;
409 };
410 
411 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
412 {
413 	if (rt == arg->net->ipv6.fib6_null_entry)
414 		return;
415 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
416 }
417 
418 static int fib6_node_dump(struct fib6_walker *w)
419 {
420 	struct fib6_info *rt;
421 
422 	for_each_fib6_walker_rt(w)
423 		fib6_rt_dump(rt, w->args);
424 	w->leaf = NULL;
425 	return 0;
426 }
427 
428 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
429 			    struct fib6_walker *w)
430 {
431 	w->root = &tb->tb6_root;
432 	spin_lock_bh(&tb->tb6_lock);
433 	fib6_walk(net, w);
434 	spin_unlock_bh(&tb->tb6_lock);
435 }
436 
437 /* Called with rcu_read_lock() */
438 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
439 {
440 	struct fib6_dump_arg arg;
441 	struct fib6_walker *w;
442 	unsigned int h;
443 
444 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
445 	if (!w)
446 		return -ENOMEM;
447 
448 	w->func = fib6_node_dump;
449 	arg.net = net;
450 	arg.nb = nb;
451 	w->args = &arg;
452 
453 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
454 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
455 		struct fib6_table *tb;
456 
457 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
458 			fib6_table_dump(net, tb, w);
459 	}
460 
461 	kfree(w);
462 
463 	return 0;
464 }
465 
466 static int fib6_dump_node(struct fib6_walker *w)
467 {
468 	int res;
469 	struct fib6_info *rt;
470 
471 	for_each_fib6_walker_rt(w) {
472 		res = rt6_dump_route(rt, w->args);
473 		if (res < 0) {
474 			/* Frame is full, suspend walking */
475 			w->leaf = rt;
476 			return 1;
477 		}
478 
479 		/* Multipath routes are dumped in one route with the
480 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
481 		 * last sibling of this route (no need to dump the
482 		 * sibling routes again)
483 		 */
484 		if (rt->fib6_nsiblings)
485 			rt = list_last_entry(&rt->fib6_siblings,
486 					     struct fib6_info,
487 					     fib6_siblings);
488 	}
489 	w->leaf = NULL;
490 	return 0;
491 }
492 
493 static void fib6_dump_end(struct netlink_callback *cb)
494 {
495 	struct net *net = sock_net(cb->skb->sk);
496 	struct fib6_walker *w = (void *)cb->args[2];
497 
498 	if (w) {
499 		if (cb->args[4]) {
500 			cb->args[4] = 0;
501 			fib6_walker_unlink(net, w);
502 		}
503 		cb->args[2] = 0;
504 		kfree(w);
505 	}
506 	cb->done = (void *)cb->args[3];
507 	cb->args[1] = 3;
508 }
509 
510 static int fib6_dump_done(struct netlink_callback *cb)
511 {
512 	fib6_dump_end(cb);
513 	return cb->done ? cb->done(cb) : 0;
514 }
515 
516 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
517 			   struct netlink_callback *cb)
518 {
519 	struct net *net = sock_net(skb->sk);
520 	struct fib6_walker *w;
521 	int res;
522 
523 	w = (void *)cb->args[2];
524 	w->root = &table->tb6_root;
525 
526 	if (cb->args[4] == 0) {
527 		w->count = 0;
528 		w->skip = 0;
529 
530 		spin_lock_bh(&table->tb6_lock);
531 		res = fib6_walk(net, w);
532 		spin_unlock_bh(&table->tb6_lock);
533 		if (res > 0) {
534 			cb->args[4] = 1;
535 			cb->args[5] = w->root->fn_sernum;
536 		}
537 	} else {
538 		if (cb->args[5] != w->root->fn_sernum) {
539 			/* Begin at the root if the tree changed */
540 			cb->args[5] = w->root->fn_sernum;
541 			w->state = FWS_INIT;
542 			w->node = w->root;
543 			w->skip = w->count;
544 		} else
545 			w->skip = 0;
546 
547 		spin_lock_bh(&table->tb6_lock);
548 		res = fib6_walk_continue(w);
549 		spin_unlock_bh(&table->tb6_lock);
550 		if (res <= 0) {
551 			fib6_walker_unlink(net, w);
552 			cb->args[4] = 0;
553 		}
554 	}
555 
556 	return res;
557 }
558 
559 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
560 {
561 	const struct nlmsghdr *nlh = cb->nlh;
562 	struct net *net = sock_net(skb->sk);
563 	struct rt6_rtnl_dump_arg arg = {};
564 	unsigned int h, s_h;
565 	unsigned int e = 0, s_e;
566 	struct fib6_walker *w;
567 	struct fib6_table *tb;
568 	struct hlist_head *head;
569 	int res = 0;
570 
571 	if (cb->strict_check) {
572 		int err;
573 
574 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
575 		if (err < 0)
576 			return err;
577 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
578 		struct rtmsg *rtm = nlmsg_data(nlh);
579 
580 		arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
581 	}
582 
583 	/* fib entries are never clones */
584 	if (arg.filter.flags & RTM_F_CLONED)
585 		goto out;
586 
587 	w = (void *)cb->args[2];
588 	if (!w) {
589 		/* New dump:
590 		 *
591 		 * 1. hook callback destructor.
592 		 */
593 		cb->args[3] = (long)cb->done;
594 		cb->done = fib6_dump_done;
595 
596 		/*
597 		 * 2. allocate and initialize walker.
598 		 */
599 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
600 		if (!w)
601 			return -ENOMEM;
602 		w->func = fib6_dump_node;
603 		cb->args[2] = (long)w;
604 	}
605 
606 	arg.skb = skb;
607 	arg.cb = cb;
608 	arg.net = net;
609 	w->args = &arg;
610 
611 	if (arg.filter.table_id) {
612 		tb = fib6_get_table(net, arg.filter.table_id);
613 		if (!tb) {
614 			if (arg.filter.dump_all_families)
615 				goto out;
616 
617 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
618 			return -ENOENT;
619 		}
620 
621 		if (!cb->args[0]) {
622 			res = fib6_dump_table(tb, skb, cb);
623 			if (!res)
624 				cb->args[0] = 1;
625 		}
626 		goto out;
627 	}
628 
629 	s_h = cb->args[0];
630 	s_e = cb->args[1];
631 
632 	rcu_read_lock();
633 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
634 		e = 0;
635 		head = &net->ipv6.fib_table_hash[h];
636 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
637 			if (e < s_e)
638 				goto next;
639 			res = fib6_dump_table(tb, skb, cb);
640 			if (res != 0)
641 				goto out_unlock;
642 next:
643 			e++;
644 		}
645 	}
646 out_unlock:
647 	rcu_read_unlock();
648 	cb->args[1] = e;
649 	cb->args[0] = h;
650 out:
651 	res = res < 0 ? res : skb->len;
652 	if (res <= 0)
653 		fib6_dump_end(cb);
654 	return res;
655 }
656 
657 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
658 {
659 	if (!f6i)
660 		return;
661 
662 	if (f6i->fib6_metrics == &dst_default_metrics) {
663 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
664 
665 		if (!p)
666 			return;
667 
668 		refcount_set(&p->refcnt, 1);
669 		f6i->fib6_metrics = p;
670 	}
671 
672 	f6i->fib6_metrics->metrics[metric - 1] = val;
673 }
674 
675 /*
676  *	Routing Table
677  *
678  *	return the appropriate node for a routing tree "add" operation
679  *	by either creating and inserting or by returning an existing
680  *	node.
681  */
682 
683 static struct fib6_node *fib6_add_1(struct net *net,
684 				    struct fib6_table *table,
685 				    struct fib6_node *root,
686 				    struct in6_addr *addr, int plen,
687 				    int offset, int allow_create,
688 				    int replace_required,
689 				    struct netlink_ext_ack *extack)
690 {
691 	struct fib6_node *fn, *in, *ln;
692 	struct fib6_node *pn = NULL;
693 	struct rt6key *key;
694 	int	bit;
695 	__be32	dir = 0;
696 
697 	RT6_TRACE("fib6_add_1\n");
698 
699 	/* insert node in tree */
700 
701 	fn = root;
702 
703 	do {
704 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
705 					    lockdep_is_held(&table->tb6_lock));
706 		key = (struct rt6key *)((u8 *)leaf + offset);
707 
708 		/*
709 		 *	Prefix match
710 		 */
711 		if (plen < fn->fn_bit ||
712 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
713 			if (!allow_create) {
714 				if (replace_required) {
715 					NL_SET_ERR_MSG(extack,
716 						       "Can not replace route - no match found");
717 					pr_warn("Can't replace route, no match found\n");
718 					return ERR_PTR(-ENOENT);
719 				}
720 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
721 			}
722 			goto insert_above;
723 		}
724 
725 		/*
726 		 *	Exact match ?
727 		 */
728 
729 		if (plen == fn->fn_bit) {
730 			/* clean up an intermediate node */
731 			if (!(fn->fn_flags & RTN_RTINFO)) {
732 				RCU_INIT_POINTER(fn->leaf, NULL);
733 				fib6_info_release(leaf);
734 			/* remove null_entry in the root node */
735 			} else if (fn->fn_flags & RTN_TL_ROOT &&
736 				   rcu_access_pointer(fn->leaf) ==
737 				   net->ipv6.fib6_null_entry) {
738 				RCU_INIT_POINTER(fn->leaf, NULL);
739 			}
740 
741 			return fn;
742 		}
743 
744 		/*
745 		 *	We have more bits to go
746 		 */
747 
748 		/* Try to walk down on tree. */
749 		dir = addr_bit_set(addr, fn->fn_bit);
750 		pn = fn;
751 		fn = dir ?
752 		     rcu_dereference_protected(fn->right,
753 					lockdep_is_held(&table->tb6_lock)) :
754 		     rcu_dereference_protected(fn->left,
755 					lockdep_is_held(&table->tb6_lock));
756 	} while (fn);
757 
758 	if (!allow_create) {
759 		/* We should not create new node because
760 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
761 		 * I assume it is safe to require NLM_F_CREATE when
762 		 * REPLACE flag is used! Later we may want to remove the
763 		 * check for replace_required, because according
764 		 * to netlink specification, NLM_F_CREATE
765 		 * MUST be specified if new route is created.
766 		 * That would keep IPv6 consistent with IPv4
767 		 */
768 		if (replace_required) {
769 			NL_SET_ERR_MSG(extack,
770 				       "Can not replace route - no match found");
771 			pr_warn("Can't replace route, no match found\n");
772 			return ERR_PTR(-ENOENT);
773 		}
774 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
775 	}
776 	/*
777 	 *	We walked to the bottom of tree.
778 	 *	Create new leaf node without children.
779 	 */
780 
781 	ln = node_alloc(net);
782 
783 	if (!ln)
784 		return ERR_PTR(-ENOMEM);
785 	ln->fn_bit = plen;
786 	RCU_INIT_POINTER(ln->parent, pn);
787 
788 	if (dir)
789 		rcu_assign_pointer(pn->right, ln);
790 	else
791 		rcu_assign_pointer(pn->left, ln);
792 
793 	return ln;
794 
795 
796 insert_above:
797 	/*
798 	 * split since we don't have a common prefix anymore or
799 	 * we have a less significant route.
800 	 * we've to insert an intermediate node on the list
801 	 * this new node will point to the one we need to create
802 	 * and the current
803 	 */
804 
805 	pn = rcu_dereference_protected(fn->parent,
806 				       lockdep_is_held(&table->tb6_lock));
807 
808 	/* find 1st bit in difference between the 2 addrs.
809 
810 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
811 	   but if it is >= plen, the value is ignored in any case.
812 	 */
813 
814 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
815 
816 	/*
817 	 *		(intermediate)[in]
818 	 *	          /	   \
819 	 *	(new leaf node)[ln] (old node)[fn]
820 	 */
821 	if (plen > bit) {
822 		in = node_alloc(net);
823 		ln = node_alloc(net);
824 
825 		if (!in || !ln) {
826 			if (in)
827 				node_free_immediate(net, in);
828 			if (ln)
829 				node_free_immediate(net, ln);
830 			return ERR_PTR(-ENOMEM);
831 		}
832 
833 		/*
834 		 * new intermediate node.
835 		 * RTN_RTINFO will
836 		 * be off since that an address that chooses one of
837 		 * the branches would not match less specific routes
838 		 * in the other branch
839 		 */
840 
841 		in->fn_bit = bit;
842 
843 		RCU_INIT_POINTER(in->parent, pn);
844 		in->leaf = fn->leaf;
845 		fib6_info_hold(rcu_dereference_protected(in->leaf,
846 				lockdep_is_held(&table->tb6_lock)));
847 
848 		/* update parent pointer */
849 		if (dir)
850 			rcu_assign_pointer(pn->right, in);
851 		else
852 			rcu_assign_pointer(pn->left, in);
853 
854 		ln->fn_bit = plen;
855 
856 		RCU_INIT_POINTER(ln->parent, in);
857 		rcu_assign_pointer(fn->parent, in);
858 
859 		if (addr_bit_set(addr, bit)) {
860 			rcu_assign_pointer(in->right, ln);
861 			rcu_assign_pointer(in->left, fn);
862 		} else {
863 			rcu_assign_pointer(in->left, ln);
864 			rcu_assign_pointer(in->right, fn);
865 		}
866 	} else { /* plen <= bit */
867 
868 		/*
869 		 *		(new leaf node)[ln]
870 		 *	          /	   \
871 		 *	     (old node)[fn] NULL
872 		 */
873 
874 		ln = node_alloc(net);
875 
876 		if (!ln)
877 			return ERR_PTR(-ENOMEM);
878 
879 		ln->fn_bit = plen;
880 
881 		RCU_INIT_POINTER(ln->parent, pn);
882 
883 		if (addr_bit_set(&key->addr, plen))
884 			RCU_INIT_POINTER(ln->right, fn);
885 		else
886 			RCU_INIT_POINTER(ln->left, fn);
887 
888 		rcu_assign_pointer(fn->parent, ln);
889 
890 		if (dir)
891 			rcu_assign_pointer(pn->right, ln);
892 		else
893 			rcu_assign_pointer(pn->left, ln);
894 	}
895 	return ln;
896 }
897 
898 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
899 				const struct fib6_table *table)
900 {
901 	int cpu;
902 
903 	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
904 	 * while we are cleaning them here.
905 	 */
906 	f6i->fib6_destroying = 1;
907 	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
908 
909 	/* release the reference to this fib entry from
910 	 * all of its cached pcpu routes
911 	 */
912 	for_each_possible_cpu(cpu) {
913 		struct rt6_info **ppcpu_rt;
914 		struct rt6_info *pcpu_rt;
915 
916 		ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
917 		pcpu_rt = *ppcpu_rt;
918 		if (pcpu_rt) {
919 			struct fib6_info *from;
920 
921 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
922 			fib6_info_release(from);
923 		}
924 	}
925 }
926 
927 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
928 			  struct net *net)
929 {
930 	struct fib6_table *table = rt->fib6_table;
931 
932 	if (rt->rt6i_pcpu)
933 		fib6_drop_pcpu_from(rt, table);
934 
935 	if (refcount_read(&rt->fib6_ref) != 1) {
936 		/* This route is used as dummy address holder in some split
937 		 * nodes. It is not leaked, but it still holds other resources,
938 		 * which must be released in time. So, scan ascendant nodes
939 		 * and replace dummy references to this route with references
940 		 * to still alive ones.
941 		 */
942 		while (fn) {
943 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
944 					    lockdep_is_held(&table->tb6_lock));
945 			struct fib6_info *new_leaf;
946 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
947 				new_leaf = fib6_find_prefix(net, table, fn);
948 				fib6_info_hold(new_leaf);
949 
950 				rcu_assign_pointer(fn->leaf, new_leaf);
951 				fib6_info_release(rt);
952 			}
953 			fn = rcu_dereference_protected(fn->parent,
954 				    lockdep_is_held(&table->tb6_lock));
955 		}
956 	}
957 }
958 
959 /*
960  *	Insert routing information in a node.
961  */
962 
963 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
964 			    struct nl_info *info,
965 			    struct netlink_ext_ack *extack)
966 {
967 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
968 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
969 	struct fib6_info *iter = NULL;
970 	struct fib6_info __rcu **ins;
971 	struct fib6_info __rcu **fallback_ins = NULL;
972 	int replace = (info->nlh &&
973 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
974 	int add = (!info->nlh ||
975 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
976 	int found = 0;
977 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
978 	u16 nlflags = NLM_F_EXCL;
979 	int err;
980 
981 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
982 		nlflags |= NLM_F_APPEND;
983 
984 	ins = &fn->leaf;
985 
986 	for (iter = leaf; iter;
987 	     iter = rcu_dereference_protected(iter->fib6_next,
988 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
989 		/*
990 		 *	Search for duplicates
991 		 */
992 
993 		if (iter->fib6_metric == rt->fib6_metric) {
994 			/*
995 			 *	Same priority level
996 			 */
997 			if (info->nlh &&
998 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
999 				return -EEXIST;
1000 
1001 			nlflags &= ~NLM_F_EXCL;
1002 			if (replace) {
1003 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1004 					found++;
1005 					break;
1006 				}
1007 				if (rt_can_ecmp)
1008 					fallback_ins = fallback_ins ?: ins;
1009 				goto next_iter;
1010 			}
1011 
1012 			if (rt6_duplicate_nexthop(iter, rt)) {
1013 				if (rt->fib6_nsiblings)
1014 					rt->fib6_nsiblings = 0;
1015 				if (!(iter->fib6_flags & RTF_EXPIRES))
1016 					return -EEXIST;
1017 				if (!(rt->fib6_flags & RTF_EXPIRES))
1018 					fib6_clean_expires(iter);
1019 				else
1020 					fib6_set_expires(iter, rt->expires);
1021 
1022 				if (rt->fib6_pmtu)
1023 					fib6_metric_set(iter, RTAX_MTU,
1024 							rt->fib6_pmtu);
1025 				return -EEXIST;
1026 			}
1027 			/* If we have the same destination and the same metric,
1028 			 * but not the same gateway, then the route we try to
1029 			 * add is sibling to this route, increment our counter
1030 			 * of siblings, and later we will add our route to the
1031 			 * list.
1032 			 * Only static routes (which don't have flag
1033 			 * RTF_EXPIRES) are used for ECMPv6.
1034 			 *
1035 			 * To avoid long list, we only had siblings if the
1036 			 * route have a gateway.
1037 			 */
1038 			if (rt_can_ecmp &&
1039 			    rt6_qualify_for_ecmp(iter))
1040 				rt->fib6_nsiblings++;
1041 		}
1042 
1043 		if (iter->fib6_metric > rt->fib6_metric)
1044 			break;
1045 
1046 next_iter:
1047 		ins = &iter->fib6_next;
1048 	}
1049 
1050 	if (fallback_ins && !found) {
1051 		/* No ECMP-able route found, replace first non-ECMP one */
1052 		ins = fallback_ins;
1053 		iter = rcu_dereference_protected(*ins,
1054 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1055 		found++;
1056 	}
1057 
1058 	/* Reset round-robin state, if necessary */
1059 	if (ins == &fn->leaf)
1060 		fn->rr_ptr = NULL;
1061 
1062 	/* Link this route to others same route. */
1063 	if (rt->fib6_nsiblings) {
1064 		unsigned int fib6_nsiblings;
1065 		struct fib6_info *sibling, *temp_sibling;
1066 
1067 		/* Find the first route that have the same metric */
1068 		sibling = leaf;
1069 		while (sibling) {
1070 			if (sibling->fib6_metric == rt->fib6_metric &&
1071 			    rt6_qualify_for_ecmp(sibling)) {
1072 				list_add_tail(&rt->fib6_siblings,
1073 					      &sibling->fib6_siblings);
1074 				break;
1075 			}
1076 			sibling = rcu_dereference_protected(sibling->fib6_next,
1077 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1078 		}
1079 		/* For each sibling in the list, increment the counter of
1080 		 * siblings. BUG() if counters does not match, list of siblings
1081 		 * is broken!
1082 		 */
1083 		fib6_nsiblings = 0;
1084 		list_for_each_entry_safe(sibling, temp_sibling,
1085 					 &rt->fib6_siblings, fib6_siblings) {
1086 			sibling->fib6_nsiblings++;
1087 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1088 			fib6_nsiblings++;
1089 		}
1090 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1091 		rt6_multipath_rebalance(temp_sibling);
1092 	}
1093 
1094 	/*
1095 	 *	insert node
1096 	 */
1097 	if (!replace) {
1098 		if (!add)
1099 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1100 
1101 add:
1102 		nlflags |= NLM_F_CREATE;
1103 
1104 		err = call_fib6_entry_notifiers(info->nl_net,
1105 						FIB_EVENT_ENTRY_ADD,
1106 						rt, extack);
1107 		if (err)
1108 			return err;
1109 
1110 		rcu_assign_pointer(rt->fib6_next, iter);
1111 		fib6_info_hold(rt);
1112 		rcu_assign_pointer(rt->fib6_node, fn);
1113 		rcu_assign_pointer(*ins, rt);
1114 		if (!info->skip_notify)
1115 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1116 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1117 
1118 		if (!(fn->fn_flags & RTN_RTINFO)) {
1119 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1120 			fn->fn_flags |= RTN_RTINFO;
1121 		}
1122 
1123 	} else {
1124 		int nsiblings;
1125 
1126 		if (!found) {
1127 			if (add)
1128 				goto add;
1129 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1130 			return -ENOENT;
1131 		}
1132 
1133 		err = call_fib6_entry_notifiers(info->nl_net,
1134 						FIB_EVENT_ENTRY_REPLACE,
1135 						rt, extack);
1136 		if (err)
1137 			return err;
1138 
1139 		fib6_info_hold(rt);
1140 		rcu_assign_pointer(rt->fib6_node, fn);
1141 		rt->fib6_next = iter->fib6_next;
1142 		rcu_assign_pointer(*ins, rt);
1143 		if (!info->skip_notify)
1144 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1145 		if (!(fn->fn_flags & RTN_RTINFO)) {
1146 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1147 			fn->fn_flags |= RTN_RTINFO;
1148 		}
1149 		nsiblings = iter->fib6_nsiblings;
1150 		iter->fib6_node = NULL;
1151 		fib6_purge_rt(iter, fn, info->nl_net);
1152 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1153 			fn->rr_ptr = NULL;
1154 		fib6_info_release(iter);
1155 
1156 		if (nsiblings) {
1157 			/* Replacing an ECMP route, remove all siblings */
1158 			ins = &rt->fib6_next;
1159 			iter = rcu_dereference_protected(*ins,
1160 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1161 			while (iter) {
1162 				if (iter->fib6_metric > rt->fib6_metric)
1163 					break;
1164 				if (rt6_qualify_for_ecmp(iter)) {
1165 					*ins = iter->fib6_next;
1166 					iter->fib6_node = NULL;
1167 					fib6_purge_rt(iter, fn, info->nl_net);
1168 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1169 						fn->rr_ptr = NULL;
1170 					fib6_info_release(iter);
1171 					nsiblings--;
1172 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1173 				} else {
1174 					ins = &iter->fib6_next;
1175 				}
1176 				iter = rcu_dereference_protected(*ins,
1177 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1178 			}
1179 			WARN_ON(nsiblings != 0);
1180 		}
1181 	}
1182 
1183 	return 0;
1184 }
1185 
1186 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1187 {
1188 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1189 	    (rt->fib6_flags & RTF_EXPIRES))
1190 		mod_timer(&net->ipv6.ip6_fib_timer,
1191 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1192 }
1193 
1194 void fib6_force_start_gc(struct net *net)
1195 {
1196 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1197 		mod_timer(&net->ipv6.ip6_fib_timer,
1198 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1199 }
1200 
1201 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1202 					   int sernum)
1203 {
1204 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1205 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1206 
1207 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1208 	smp_wmb();
1209 	while (fn) {
1210 		fn->fn_sernum = sernum;
1211 		fn = rcu_dereference_protected(fn->parent,
1212 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1213 	}
1214 }
1215 
1216 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1217 {
1218 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1219 }
1220 
1221 /*
1222  *	Add routing information to the routing tree.
1223  *	<destination addr>/<source addr>
1224  *	with source addr info in sub-trees
1225  *	Need to own table->tb6_lock
1226  */
1227 
1228 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1229 	     struct nl_info *info, struct netlink_ext_ack *extack)
1230 {
1231 	struct fib6_table *table = rt->fib6_table;
1232 	struct fib6_node *fn, *pn = NULL;
1233 	int err = -ENOMEM;
1234 	int allow_create = 1;
1235 	int replace_required = 0;
1236 	int sernum = fib6_new_sernum(info->nl_net);
1237 
1238 	if (info->nlh) {
1239 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1240 			allow_create = 0;
1241 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1242 			replace_required = 1;
1243 	}
1244 	if (!allow_create && !replace_required)
1245 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1246 
1247 	fn = fib6_add_1(info->nl_net, table, root,
1248 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1249 			offsetof(struct fib6_info, fib6_dst), allow_create,
1250 			replace_required, extack);
1251 	if (IS_ERR(fn)) {
1252 		err = PTR_ERR(fn);
1253 		fn = NULL;
1254 		goto out;
1255 	}
1256 
1257 	pn = fn;
1258 
1259 #ifdef CONFIG_IPV6_SUBTREES
1260 	if (rt->fib6_src.plen) {
1261 		struct fib6_node *sn;
1262 
1263 		if (!rcu_access_pointer(fn->subtree)) {
1264 			struct fib6_node *sfn;
1265 
1266 			/*
1267 			 * Create subtree.
1268 			 *
1269 			 *		fn[main tree]
1270 			 *		|
1271 			 *		sfn[subtree root]
1272 			 *		   \
1273 			 *		    sn[new leaf node]
1274 			 */
1275 
1276 			/* Create subtree root node */
1277 			sfn = node_alloc(info->nl_net);
1278 			if (!sfn)
1279 				goto failure;
1280 
1281 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1282 			rcu_assign_pointer(sfn->leaf,
1283 					   info->nl_net->ipv6.fib6_null_entry);
1284 			sfn->fn_flags = RTN_ROOT;
1285 
1286 			/* Now add the first leaf node to new subtree */
1287 
1288 			sn = fib6_add_1(info->nl_net, table, sfn,
1289 					&rt->fib6_src.addr, rt->fib6_src.plen,
1290 					offsetof(struct fib6_info, fib6_src),
1291 					allow_create, replace_required, extack);
1292 
1293 			if (IS_ERR(sn)) {
1294 				/* If it is failed, discard just allocated
1295 				   root, and then (in failure) stale node
1296 				   in main tree.
1297 				 */
1298 				node_free_immediate(info->nl_net, sfn);
1299 				err = PTR_ERR(sn);
1300 				goto failure;
1301 			}
1302 
1303 			/* Now link new subtree to main tree */
1304 			rcu_assign_pointer(sfn->parent, fn);
1305 			rcu_assign_pointer(fn->subtree, sfn);
1306 		} else {
1307 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1308 					&rt->fib6_src.addr, rt->fib6_src.plen,
1309 					offsetof(struct fib6_info, fib6_src),
1310 					allow_create, replace_required, extack);
1311 
1312 			if (IS_ERR(sn)) {
1313 				err = PTR_ERR(sn);
1314 				goto failure;
1315 			}
1316 		}
1317 
1318 		if (!rcu_access_pointer(fn->leaf)) {
1319 			if (fn->fn_flags & RTN_TL_ROOT) {
1320 				/* put back null_entry for root node */
1321 				rcu_assign_pointer(fn->leaf,
1322 					    info->nl_net->ipv6.fib6_null_entry);
1323 			} else {
1324 				fib6_info_hold(rt);
1325 				rcu_assign_pointer(fn->leaf, rt);
1326 			}
1327 		}
1328 		fn = sn;
1329 	}
1330 #endif
1331 
1332 	err = fib6_add_rt2node(fn, rt, info, extack);
1333 	if (!err) {
1334 		__fib6_update_sernum_upto_root(rt, sernum);
1335 		fib6_start_gc(info->nl_net, rt);
1336 	}
1337 
1338 out:
1339 	if (err) {
1340 #ifdef CONFIG_IPV6_SUBTREES
1341 		/*
1342 		 * If fib6_add_1 has cleared the old leaf pointer in the
1343 		 * super-tree leaf node we have to find a new one for it.
1344 		 */
1345 		if (pn != fn) {
1346 			struct fib6_info *pn_leaf =
1347 				rcu_dereference_protected(pn->leaf,
1348 				    lockdep_is_held(&table->tb6_lock));
1349 			if (pn_leaf == rt) {
1350 				pn_leaf = NULL;
1351 				RCU_INIT_POINTER(pn->leaf, NULL);
1352 				fib6_info_release(rt);
1353 			}
1354 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1355 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1356 							   pn);
1357 #if RT6_DEBUG >= 2
1358 				if (!pn_leaf) {
1359 					WARN_ON(!pn_leaf);
1360 					pn_leaf =
1361 					    info->nl_net->ipv6.fib6_null_entry;
1362 				}
1363 #endif
1364 				fib6_info_hold(pn_leaf);
1365 				rcu_assign_pointer(pn->leaf, pn_leaf);
1366 			}
1367 		}
1368 #endif
1369 		goto failure;
1370 	}
1371 	return err;
1372 
1373 failure:
1374 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1375 	 * 1. fn is an intermediate node and we failed to add the new
1376 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1377 	 * failure case.
1378 	 * 2. fn is the root node in the table and we fail to add the first
1379 	 * default route to it.
1380 	 */
1381 	if (fn &&
1382 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1383 	     (fn->fn_flags & RTN_TL_ROOT &&
1384 	      !rcu_access_pointer(fn->leaf))))
1385 		fib6_repair_tree(info->nl_net, table, fn);
1386 	return err;
1387 }
1388 
1389 /*
1390  *	Routing tree lookup
1391  *
1392  */
1393 
1394 struct lookup_args {
1395 	int			offset;		/* key offset on fib6_info */
1396 	const struct in6_addr	*addr;		/* search key			*/
1397 };
1398 
1399 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1400 					    struct lookup_args *args)
1401 {
1402 	struct fib6_node *fn;
1403 	__be32 dir;
1404 
1405 	if (unlikely(args->offset == 0))
1406 		return NULL;
1407 
1408 	/*
1409 	 *	Descend on a tree
1410 	 */
1411 
1412 	fn = root;
1413 
1414 	for (;;) {
1415 		struct fib6_node *next;
1416 
1417 		dir = addr_bit_set(args->addr, fn->fn_bit);
1418 
1419 		next = dir ? rcu_dereference(fn->right) :
1420 			     rcu_dereference(fn->left);
1421 
1422 		if (next) {
1423 			fn = next;
1424 			continue;
1425 		}
1426 		break;
1427 	}
1428 
1429 	while (fn) {
1430 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1431 
1432 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1433 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1434 			struct rt6key *key;
1435 
1436 			if (!leaf)
1437 				goto backtrack;
1438 
1439 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1440 
1441 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1442 #ifdef CONFIG_IPV6_SUBTREES
1443 				if (subtree) {
1444 					struct fib6_node *sfn;
1445 					sfn = fib6_node_lookup_1(subtree,
1446 								 args + 1);
1447 					if (!sfn)
1448 						goto backtrack;
1449 					fn = sfn;
1450 				}
1451 #endif
1452 				if (fn->fn_flags & RTN_RTINFO)
1453 					return fn;
1454 			}
1455 		}
1456 backtrack:
1457 		if (fn->fn_flags & RTN_ROOT)
1458 			break;
1459 
1460 		fn = rcu_dereference(fn->parent);
1461 	}
1462 
1463 	return NULL;
1464 }
1465 
1466 /* called with rcu_read_lock() held
1467  */
1468 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1469 				   const struct in6_addr *daddr,
1470 				   const struct in6_addr *saddr)
1471 {
1472 	struct fib6_node *fn;
1473 	struct lookup_args args[] = {
1474 		{
1475 			.offset = offsetof(struct fib6_info, fib6_dst),
1476 			.addr = daddr,
1477 		},
1478 #ifdef CONFIG_IPV6_SUBTREES
1479 		{
1480 			.offset = offsetof(struct fib6_info, fib6_src),
1481 			.addr = saddr,
1482 		},
1483 #endif
1484 		{
1485 			.offset = 0,	/* sentinel */
1486 		}
1487 	};
1488 
1489 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1490 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1491 		fn = root;
1492 
1493 	return fn;
1494 }
1495 
1496 /*
1497  *	Get node with specified destination prefix (and source prefix,
1498  *	if subtrees are used)
1499  *	exact_match == true means we try to find fn with exact match of
1500  *	the passed in prefix addr
1501  *	exact_match == false means we try to find fn with longest prefix
1502  *	match of the passed in prefix addr. This is useful for finding fn
1503  *	for cached route as it will be stored in the exception table under
1504  *	the node with longest prefix length.
1505  */
1506 
1507 
1508 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1509 				       const struct in6_addr *addr,
1510 				       int plen, int offset,
1511 				       bool exact_match)
1512 {
1513 	struct fib6_node *fn, *prev = NULL;
1514 
1515 	for (fn = root; fn ; ) {
1516 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1517 		struct rt6key *key;
1518 
1519 		/* This node is being deleted */
1520 		if (!leaf) {
1521 			if (plen <= fn->fn_bit)
1522 				goto out;
1523 			else
1524 				goto next;
1525 		}
1526 
1527 		key = (struct rt6key *)((u8 *)leaf + offset);
1528 
1529 		/*
1530 		 *	Prefix match
1531 		 */
1532 		if (plen < fn->fn_bit ||
1533 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1534 			goto out;
1535 
1536 		if (plen == fn->fn_bit)
1537 			return fn;
1538 
1539 		prev = fn;
1540 
1541 next:
1542 		/*
1543 		 *	We have more bits to go
1544 		 */
1545 		if (addr_bit_set(addr, fn->fn_bit))
1546 			fn = rcu_dereference(fn->right);
1547 		else
1548 			fn = rcu_dereference(fn->left);
1549 	}
1550 out:
1551 	if (exact_match)
1552 		return NULL;
1553 	else
1554 		return prev;
1555 }
1556 
1557 struct fib6_node *fib6_locate(struct fib6_node *root,
1558 			      const struct in6_addr *daddr, int dst_len,
1559 			      const struct in6_addr *saddr, int src_len,
1560 			      bool exact_match)
1561 {
1562 	struct fib6_node *fn;
1563 
1564 	fn = fib6_locate_1(root, daddr, dst_len,
1565 			   offsetof(struct fib6_info, fib6_dst),
1566 			   exact_match);
1567 
1568 #ifdef CONFIG_IPV6_SUBTREES
1569 	if (src_len) {
1570 		WARN_ON(saddr == NULL);
1571 		if (fn) {
1572 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1573 
1574 			if (subtree) {
1575 				fn = fib6_locate_1(subtree, saddr, src_len,
1576 					   offsetof(struct fib6_info, fib6_src),
1577 					   exact_match);
1578 			}
1579 		}
1580 	}
1581 #endif
1582 
1583 	if (fn && fn->fn_flags & RTN_RTINFO)
1584 		return fn;
1585 
1586 	return NULL;
1587 }
1588 
1589 
1590 /*
1591  *	Deletion
1592  *
1593  */
1594 
1595 static struct fib6_info *fib6_find_prefix(struct net *net,
1596 					 struct fib6_table *table,
1597 					 struct fib6_node *fn)
1598 {
1599 	struct fib6_node *child_left, *child_right;
1600 
1601 	if (fn->fn_flags & RTN_ROOT)
1602 		return net->ipv6.fib6_null_entry;
1603 
1604 	while (fn) {
1605 		child_left = rcu_dereference_protected(fn->left,
1606 				    lockdep_is_held(&table->tb6_lock));
1607 		child_right = rcu_dereference_protected(fn->right,
1608 				    lockdep_is_held(&table->tb6_lock));
1609 		if (child_left)
1610 			return rcu_dereference_protected(child_left->leaf,
1611 					lockdep_is_held(&table->tb6_lock));
1612 		if (child_right)
1613 			return rcu_dereference_protected(child_right->leaf,
1614 					lockdep_is_held(&table->tb6_lock));
1615 
1616 		fn = FIB6_SUBTREE(fn);
1617 	}
1618 	return NULL;
1619 }
1620 
1621 /*
1622  *	Called to trim the tree of intermediate nodes when possible. "fn"
1623  *	is the node we want to try and remove.
1624  *	Need to own table->tb6_lock
1625  */
1626 
1627 static struct fib6_node *fib6_repair_tree(struct net *net,
1628 					  struct fib6_table *table,
1629 					  struct fib6_node *fn)
1630 {
1631 	int children;
1632 	int nstate;
1633 	struct fib6_node *child;
1634 	struct fib6_walker *w;
1635 	int iter = 0;
1636 
1637 	/* Set fn->leaf to null_entry for root node. */
1638 	if (fn->fn_flags & RTN_TL_ROOT) {
1639 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1640 		return fn;
1641 	}
1642 
1643 	for (;;) {
1644 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1645 					    lockdep_is_held(&table->tb6_lock));
1646 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1647 					    lockdep_is_held(&table->tb6_lock));
1648 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1649 					    lockdep_is_held(&table->tb6_lock));
1650 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1651 					    lockdep_is_held(&table->tb6_lock));
1652 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1653 					    lockdep_is_held(&table->tb6_lock));
1654 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1655 					    lockdep_is_held(&table->tb6_lock));
1656 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1657 					    lockdep_is_held(&table->tb6_lock));
1658 		struct fib6_info *new_fn_leaf;
1659 
1660 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1661 		iter++;
1662 
1663 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1664 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1665 		WARN_ON(fn_leaf);
1666 
1667 		children = 0;
1668 		child = NULL;
1669 		if (fn_r)
1670 			child = fn_r, children |= 1;
1671 		if (fn_l)
1672 			child = fn_l, children |= 2;
1673 
1674 		if (children == 3 || FIB6_SUBTREE(fn)
1675 #ifdef CONFIG_IPV6_SUBTREES
1676 		    /* Subtree root (i.e. fn) may have one child */
1677 		    || (children && fn->fn_flags & RTN_ROOT)
1678 #endif
1679 		    ) {
1680 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1681 #if RT6_DEBUG >= 2
1682 			if (!new_fn_leaf) {
1683 				WARN_ON(!new_fn_leaf);
1684 				new_fn_leaf = net->ipv6.fib6_null_entry;
1685 			}
1686 #endif
1687 			fib6_info_hold(new_fn_leaf);
1688 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1689 			return pn;
1690 		}
1691 
1692 #ifdef CONFIG_IPV6_SUBTREES
1693 		if (FIB6_SUBTREE(pn) == fn) {
1694 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1695 			RCU_INIT_POINTER(pn->subtree, NULL);
1696 			nstate = FWS_L;
1697 		} else {
1698 			WARN_ON(fn->fn_flags & RTN_ROOT);
1699 #endif
1700 			if (pn_r == fn)
1701 				rcu_assign_pointer(pn->right, child);
1702 			else if (pn_l == fn)
1703 				rcu_assign_pointer(pn->left, child);
1704 #if RT6_DEBUG >= 2
1705 			else
1706 				WARN_ON(1);
1707 #endif
1708 			if (child)
1709 				rcu_assign_pointer(child->parent, pn);
1710 			nstate = FWS_R;
1711 #ifdef CONFIG_IPV6_SUBTREES
1712 		}
1713 #endif
1714 
1715 		read_lock(&net->ipv6.fib6_walker_lock);
1716 		FOR_WALKERS(net, w) {
1717 			if (!child) {
1718 				if (w->node == fn) {
1719 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1720 					w->node = pn;
1721 					w->state = nstate;
1722 				}
1723 			} else {
1724 				if (w->node == fn) {
1725 					w->node = child;
1726 					if (children&2) {
1727 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1728 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1729 					} else {
1730 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1731 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1732 					}
1733 				}
1734 			}
1735 		}
1736 		read_unlock(&net->ipv6.fib6_walker_lock);
1737 
1738 		node_free(net, fn);
1739 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1740 			return pn;
1741 
1742 		RCU_INIT_POINTER(pn->leaf, NULL);
1743 		fib6_info_release(pn_leaf);
1744 		fn = pn;
1745 	}
1746 }
1747 
1748 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1749 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1750 {
1751 	struct fib6_walker *w;
1752 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1753 				    lockdep_is_held(&table->tb6_lock));
1754 	struct net *net = info->nl_net;
1755 
1756 	RT6_TRACE("fib6_del_route\n");
1757 
1758 	/* Unlink it */
1759 	*rtp = rt->fib6_next;
1760 	rt->fib6_node = NULL;
1761 	net->ipv6.rt6_stats->fib_rt_entries--;
1762 	net->ipv6.rt6_stats->fib_discarded_routes++;
1763 
1764 	/* Flush all cached dst in exception table */
1765 	rt6_flush_exceptions(rt);
1766 
1767 	/* Reset round-robin state, if necessary */
1768 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1769 		fn->rr_ptr = NULL;
1770 
1771 	/* Remove this entry from other siblings */
1772 	if (rt->fib6_nsiblings) {
1773 		struct fib6_info *sibling, *next_sibling;
1774 
1775 		list_for_each_entry_safe(sibling, next_sibling,
1776 					 &rt->fib6_siblings, fib6_siblings)
1777 			sibling->fib6_nsiblings--;
1778 		rt->fib6_nsiblings = 0;
1779 		list_del_init(&rt->fib6_siblings);
1780 		rt6_multipath_rebalance(next_sibling);
1781 	}
1782 
1783 	/* Adjust walkers */
1784 	read_lock(&net->ipv6.fib6_walker_lock);
1785 	FOR_WALKERS(net, w) {
1786 		if (w->state == FWS_C && w->leaf == rt) {
1787 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1788 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1789 					    lockdep_is_held(&table->tb6_lock));
1790 			if (!w->leaf)
1791 				w->state = FWS_U;
1792 		}
1793 	}
1794 	read_unlock(&net->ipv6.fib6_walker_lock);
1795 
1796 	/* If it was last route, call fib6_repair_tree() to:
1797 	 * 1. For root node, put back null_entry as how the table was created.
1798 	 * 2. For other nodes, expunge its radix tree node.
1799 	 */
1800 	if (!rcu_access_pointer(fn->leaf)) {
1801 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1802 			fn->fn_flags &= ~RTN_RTINFO;
1803 			net->ipv6.rt6_stats->fib_route_nodes--;
1804 		}
1805 		fn = fib6_repair_tree(net, table, fn);
1806 	}
1807 
1808 	fib6_purge_rt(rt, fn, net);
1809 
1810 	call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1811 	if (!info->skip_notify)
1812 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1813 	fib6_info_release(rt);
1814 }
1815 
1816 /* Need to own table->tb6_lock */
1817 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1818 {
1819 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1820 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1821 	struct fib6_table *table = rt->fib6_table;
1822 	struct net *net = info->nl_net;
1823 	struct fib6_info __rcu **rtp;
1824 	struct fib6_info __rcu **rtp_next;
1825 
1826 	if (!fn || rt == net->ipv6.fib6_null_entry)
1827 		return -ENOENT;
1828 
1829 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1830 
1831 	/*
1832 	 *	Walk the leaf entries looking for ourself
1833 	 */
1834 
1835 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1836 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1837 					lockdep_is_held(&table->tb6_lock));
1838 		if (rt == cur) {
1839 			fib6_del_route(table, fn, rtp, info);
1840 			return 0;
1841 		}
1842 		rtp_next = &cur->fib6_next;
1843 	}
1844 	return -ENOENT;
1845 }
1846 
1847 /*
1848  *	Tree traversal function.
1849  *
1850  *	Certainly, it is not interrupt safe.
1851  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1852  *	It means, that we can modify tree during walking
1853  *	and use this function for garbage collection, clone pruning,
1854  *	cleaning tree when a device goes down etc. etc.
1855  *
1856  *	It guarantees that every node will be traversed,
1857  *	and that it will be traversed only once.
1858  *
1859  *	Callback function w->func may return:
1860  *	0 -> continue walking.
1861  *	positive value -> walking is suspended (used by tree dumps,
1862  *	and probably by gc, if it will be split to several slices)
1863  *	negative value -> terminate walking.
1864  *
1865  *	The function itself returns:
1866  *	0   -> walk is complete.
1867  *	>0  -> walk is incomplete (i.e. suspended)
1868  *	<0  -> walk is terminated by an error.
1869  *
1870  *	This function is called with tb6_lock held.
1871  */
1872 
1873 static int fib6_walk_continue(struct fib6_walker *w)
1874 {
1875 	struct fib6_node *fn, *pn, *left, *right;
1876 
1877 	/* w->root should always be table->tb6_root */
1878 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1879 
1880 	for (;;) {
1881 		fn = w->node;
1882 		if (!fn)
1883 			return 0;
1884 
1885 		switch (w->state) {
1886 #ifdef CONFIG_IPV6_SUBTREES
1887 		case FWS_S:
1888 			if (FIB6_SUBTREE(fn)) {
1889 				w->node = FIB6_SUBTREE(fn);
1890 				continue;
1891 			}
1892 			w->state = FWS_L;
1893 #endif
1894 			/* fall through */
1895 		case FWS_L:
1896 			left = rcu_dereference_protected(fn->left, 1);
1897 			if (left) {
1898 				w->node = left;
1899 				w->state = FWS_INIT;
1900 				continue;
1901 			}
1902 			w->state = FWS_R;
1903 			/* fall through */
1904 		case FWS_R:
1905 			right = rcu_dereference_protected(fn->right, 1);
1906 			if (right) {
1907 				w->node = right;
1908 				w->state = FWS_INIT;
1909 				continue;
1910 			}
1911 			w->state = FWS_C;
1912 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1913 			/* fall through */
1914 		case FWS_C:
1915 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1916 				int err;
1917 
1918 				if (w->skip) {
1919 					w->skip--;
1920 					goto skip;
1921 				}
1922 
1923 				err = w->func(w);
1924 				if (err)
1925 					return err;
1926 
1927 				w->count++;
1928 				continue;
1929 			}
1930 skip:
1931 			w->state = FWS_U;
1932 			/* fall through */
1933 		case FWS_U:
1934 			if (fn == w->root)
1935 				return 0;
1936 			pn = rcu_dereference_protected(fn->parent, 1);
1937 			left = rcu_dereference_protected(pn->left, 1);
1938 			right = rcu_dereference_protected(pn->right, 1);
1939 			w->node = pn;
1940 #ifdef CONFIG_IPV6_SUBTREES
1941 			if (FIB6_SUBTREE(pn) == fn) {
1942 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1943 				w->state = FWS_L;
1944 				continue;
1945 			}
1946 #endif
1947 			if (left == fn) {
1948 				w->state = FWS_R;
1949 				continue;
1950 			}
1951 			if (right == fn) {
1952 				w->state = FWS_C;
1953 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1954 				continue;
1955 			}
1956 #if RT6_DEBUG >= 2
1957 			WARN_ON(1);
1958 #endif
1959 		}
1960 	}
1961 }
1962 
1963 static int fib6_walk(struct net *net, struct fib6_walker *w)
1964 {
1965 	int res;
1966 
1967 	w->state = FWS_INIT;
1968 	w->node = w->root;
1969 
1970 	fib6_walker_link(net, w);
1971 	res = fib6_walk_continue(w);
1972 	if (res <= 0)
1973 		fib6_walker_unlink(net, w);
1974 	return res;
1975 }
1976 
1977 static int fib6_clean_node(struct fib6_walker *w)
1978 {
1979 	int res;
1980 	struct fib6_info *rt;
1981 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1982 	struct nl_info info = {
1983 		.nl_net = c->net,
1984 		.skip_notify = c->skip_notify,
1985 	};
1986 
1987 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1988 	    w->node->fn_sernum != c->sernum)
1989 		w->node->fn_sernum = c->sernum;
1990 
1991 	if (!c->func) {
1992 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1993 		w->leaf = NULL;
1994 		return 0;
1995 	}
1996 
1997 	for_each_fib6_walker_rt(w) {
1998 		res = c->func(rt, c->arg);
1999 		if (res == -1) {
2000 			w->leaf = rt;
2001 			res = fib6_del(rt, &info);
2002 			if (res) {
2003 #if RT6_DEBUG >= 2
2004 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2005 					 __func__, rt,
2006 					 rcu_access_pointer(rt->fib6_node),
2007 					 res);
2008 #endif
2009 				continue;
2010 			}
2011 			return 0;
2012 		} else if (res == -2) {
2013 			if (WARN_ON(!rt->fib6_nsiblings))
2014 				continue;
2015 			rt = list_last_entry(&rt->fib6_siblings,
2016 					     struct fib6_info, fib6_siblings);
2017 			continue;
2018 		}
2019 		WARN_ON(res != 0);
2020 	}
2021 	w->leaf = rt;
2022 	return 0;
2023 }
2024 
2025 /*
2026  *	Convenient frontend to tree walker.
2027  *
2028  *	func is called on each route.
2029  *		It may return -2 -> skip multipath route.
2030  *			      -1 -> delete this route.
2031  *		              0  -> continue walking
2032  */
2033 
2034 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2035 			    int (*func)(struct fib6_info *, void *arg),
2036 			    int sernum, void *arg, bool skip_notify)
2037 {
2038 	struct fib6_cleaner c;
2039 
2040 	c.w.root = root;
2041 	c.w.func = fib6_clean_node;
2042 	c.w.count = 0;
2043 	c.w.skip = 0;
2044 	c.func = func;
2045 	c.sernum = sernum;
2046 	c.arg = arg;
2047 	c.net = net;
2048 	c.skip_notify = skip_notify;
2049 
2050 	fib6_walk(net, &c.w);
2051 }
2052 
2053 static void __fib6_clean_all(struct net *net,
2054 			     int (*func)(struct fib6_info *, void *),
2055 			     int sernum, void *arg, bool skip_notify)
2056 {
2057 	struct fib6_table *table;
2058 	struct hlist_head *head;
2059 	unsigned int h;
2060 
2061 	rcu_read_lock();
2062 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2063 		head = &net->ipv6.fib_table_hash[h];
2064 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2065 			spin_lock_bh(&table->tb6_lock);
2066 			fib6_clean_tree(net, &table->tb6_root,
2067 					func, sernum, arg, skip_notify);
2068 			spin_unlock_bh(&table->tb6_lock);
2069 		}
2070 	}
2071 	rcu_read_unlock();
2072 }
2073 
2074 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2075 		    void *arg)
2076 {
2077 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2078 }
2079 
2080 void fib6_clean_all_skip_notify(struct net *net,
2081 				int (*func)(struct fib6_info *, void *),
2082 				void *arg)
2083 {
2084 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2085 }
2086 
2087 static void fib6_flush_trees(struct net *net)
2088 {
2089 	int new_sernum = fib6_new_sernum(net);
2090 
2091 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2092 }
2093 
2094 /*
2095  *	Garbage collection
2096  */
2097 
2098 static int fib6_age(struct fib6_info *rt, void *arg)
2099 {
2100 	struct fib6_gc_args *gc_args = arg;
2101 	unsigned long now = jiffies;
2102 
2103 	/*
2104 	 *	check addrconf expiration here.
2105 	 *	Routes are expired even if they are in use.
2106 	 */
2107 
2108 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2109 		if (time_after(now, rt->expires)) {
2110 			RT6_TRACE("expiring %p\n", rt);
2111 			return -1;
2112 		}
2113 		gc_args->more++;
2114 	}
2115 
2116 	/*	Also age clones in the exception table.
2117 	 *	Note, that clones are aged out
2118 	 *	only if they are not in use now.
2119 	 */
2120 	rt6_age_exceptions(rt, gc_args, now);
2121 
2122 	return 0;
2123 }
2124 
2125 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2126 {
2127 	struct fib6_gc_args gc_args;
2128 	unsigned long now;
2129 
2130 	if (force) {
2131 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2132 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2133 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2134 		return;
2135 	}
2136 	gc_args.timeout = expires ? (int)expires :
2137 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2138 	gc_args.more = 0;
2139 
2140 	fib6_clean_all(net, fib6_age, &gc_args);
2141 	now = jiffies;
2142 	net->ipv6.ip6_rt_last_gc = now;
2143 
2144 	if (gc_args.more)
2145 		mod_timer(&net->ipv6.ip6_fib_timer,
2146 			  round_jiffies(now
2147 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2148 	else
2149 		del_timer(&net->ipv6.ip6_fib_timer);
2150 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2151 }
2152 
2153 static void fib6_gc_timer_cb(struct timer_list *t)
2154 {
2155 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2156 
2157 	fib6_run_gc(0, arg, true);
2158 }
2159 
2160 static int __net_init fib6_net_init(struct net *net)
2161 {
2162 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2163 	int err;
2164 
2165 	err = fib6_notifier_init(net);
2166 	if (err)
2167 		return err;
2168 
2169 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2170 	rwlock_init(&net->ipv6.fib6_walker_lock);
2171 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2172 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2173 
2174 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2175 	if (!net->ipv6.rt6_stats)
2176 		goto out_timer;
2177 
2178 	/* Avoid false sharing : Use at least a full cache line */
2179 	size = max_t(size_t, size, L1_CACHE_BYTES);
2180 
2181 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2182 	if (!net->ipv6.fib_table_hash)
2183 		goto out_rt6_stats;
2184 
2185 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2186 					  GFP_KERNEL);
2187 	if (!net->ipv6.fib6_main_tbl)
2188 		goto out_fib_table_hash;
2189 
2190 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2191 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2192 			   net->ipv6.fib6_null_entry);
2193 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2194 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2195 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2196 
2197 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2198 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2199 					   GFP_KERNEL);
2200 	if (!net->ipv6.fib6_local_tbl)
2201 		goto out_fib6_main_tbl;
2202 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2203 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2204 			   net->ipv6.fib6_null_entry);
2205 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2206 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2207 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2208 #endif
2209 	fib6_tables_init(net);
2210 
2211 	return 0;
2212 
2213 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2214 out_fib6_main_tbl:
2215 	kfree(net->ipv6.fib6_main_tbl);
2216 #endif
2217 out_fib_table_hash:
2218 	kfree(net->ipv6.fib_table_hash);
2219 out_rt6_stats:
2220 	kfree(net->ipv6.rt6_stats);
2221 out_timer:
2222 	fib6_notifier_exit(net);
2223 	return -ENOMEM;
2224 }
2225 
2226 static void fib6_net_exit(struct net *net)
2227 {
2228 	unsigned int i;
2229 
2230 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2231 
2232 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2233 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2234 		struct hlist_node *tmp;
2235 		struct fib6_table *tb;
2236 
2237 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2238 			hlist_del(&tb->tb6_hlist);
2239 			fib6_free_table(tb);
2240 		}
2241 	}
2242 
2243 	kfree(net->ipv6.fib_table_hash);
2244 	kfree(net->ipv6.rt6_stats);
2245 	fib6_notifier_exit(net);
2246 }
2247 
2248 static struct pernet_operations fib6_net_ops = {
2249 	.init = fib6_net_init,
2250 	.exit = fib6_net_exit,
2251 };
2252 
2253 int __init fib6_init(void)
2254 {
2255 	int ret = -ENOMEM;
2256 
2257 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2258 					   sizeof(struct fib6_node),
2259 					   0, SLAB_HWCACHE_ALIGN,
2260 					   NULL);
2261 	if (!fib6_node_kmem)
2262 		goto out;
2263 
2264 	ret = register_pernet_subsys(&fib6_net_ops);
2265 	if (ret)
2266 		goto out_kmem_cache_create;
2267 
2268 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2269 				   inet6_dump_fib, 0);
2270 	if (ret)
2271 		goto out_unregister_subsys;
2272 
2273 	__fib6_flush_trees = fib6_flush_trees;
2274 out:
2275 	return ret;
2276 
2277 out_unregister_subsys:
2278 	unregister_pernet_subsys(&fib6_net_ops);
2279 out_kmem_cache_create:
2280 	kmem_cache_destroy(fib6_node_kmem);
2281 	goto out;
2282 }
2283 
2284 void fib6_gc_cleanup(void)
2285 {
2286 	unregister_pernet_subsys(&fib6_net_ops);
2287 	kmem_cache_destroy(fib6_node_kmem);
2288 }
2289 
2290 #ifdef CONFIG_PROC_FS
2291 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2292 {
2293 	struct fib6_info *rt = v;
2294 	struct ipv6_route_iter *iter = seq->private;
2295 	unsigned int flags = rt->fib6_flags;
2296 	const struct net_device *dev;
2297 
2298 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2299 
2300 #ifdef CONFIG_IPV6_SUBTREES
2301 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2302 #else
2303 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2304 #endif
2305 	if (rt->fib6_nh.fib_nh_gw_family) {
2306 		flags |= RTF_GATEWAY;
2307 		seq_printf(seq, "%pi6", &rt->fib6_nh.fib_nh_gw6);
2308 	} else {
2309 		seq_puts(seq, "00000000000000000000000000000000");
2310 	}
2311 
2312 	dev = rt->fib6_nh.fib_nh_dev;
2313 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2314 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2315 		   flags, dev ? dev->name : "");
2316 	iter->w.leaf = NULL;
2317 	return 0;
2318 }
2319 
2320 static int ipv6_route_yield(struct fib6_walker *w)
2321 {
2322 	struct ipv6_route_iter *iter = w->args;
2323 
2324 	if (!iter->skip)
2325 		return 1;
2326 
2327 	do {
2328 		iter->w.leaf = rcu_dereference_protected(
2329 				iter->w.leaf->fib6_next,
2330 				lockdep_is_held(&iter->tbl->tb6_lock));
2331 		iter->skip--;
2332 		if (!iter->skip && iter->w.leaf)
2333 			return 1;
2334 	} while (iter->w.leaf);
2335 
2336 	return 0;
2337 }
2338 
2339 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2340 				      struct net *net)
2341 {
2342 	memset(&iter->w, 0, sizeof(iter->w));
2343 	iter->w.func = ipv6_route_yield;
2344 	iter->w.root = &iter->tbl->tb6_root;
2345 	iter->w.state = FWS_INIT;
2346 	iter->w.node = iter->w.root;
2347 	iter->w.args = iter;
2348 	iter->sernum = iter->w.root->fn_sernum;
2349 	INIT_LIST_HEAD(&iter->w.lh);
2350 	fib6_walker_link(net, &iter->w);
2351 }
2352 
2353 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2354 						    struct net *net)
2355 {
2356 	unsigned int h;
2357 	struct hlist_node *node;
2358 
2359 	if (tbl) {
2360 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2361 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2362 	} else {
2363 		h = 0;
2364 		node = NULL;
2365 	}
2366 
2367 	while (!node && h < FIB6_TABLE_HASHSZ) {
2368 		node = rcu_dereference_bh(
2369 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2370 	}
2371 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2372 }
2373 
2374 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2375 {
2376 	if (iter->sernum != iter->w.root->fn_sernum) {
2377 		iter->sernum = iter->w.root->fn_sernum;
2378 		iter->w.state = FWS_INIT;
2379 		iter->w.node = iter->w.root;
2380 		WARN_ON(iter->w.skip);
2381 		iter->w.skip = iter->w.count;
2382 	}
2383 }
2384 
2385 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2386 {
2387 	int r;
2388 	struct fib6_info *n;
2389 	struct net *net = seq_file_net(seq);
2390 	struct ipv6_route_iter *iter = seq->private;
2391 
2392 	if (!v)
2393 		goto iter_table;
2394 
2395 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2396 	if (n) {
2397 		++*pos;
2398 		return n;
2399 	}
2400 
2401 iter_table:
2402 	ipv6_route_check_sernum(iter);
2403 	spin_lock_bh(&iter->tbl->tb6_lock);
2404 	r = fib6_walk_continue(&iter->w);
2405 	spin_unlock_bh(&iter->tbl->tb6_lock);
2406 	if (r > 0) {
2407 		if (v)
2408 			++*pos;
2409 		return iter->w.leaf;
2410 	} else if (r < 0) {
2411 		fib6_walker_unlink(net, &iter->w);
2412 		return NULL;
2413 	}
2414 	fib6_walker_unlink(net, &iter->w);
2415 
2416 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2417 	if (!iter->tbl)
2418 		return NULL;
2419 
2420 	ipv6_route_seq_setup_walk(iter, net);
2421 	goto iter_table;
2422 }
2423 
2424 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2425 	__acquires(RCU_BH)
2426 {
2427 	struct net *net = seq_file_net(seq);
2428 	struct ipv6_route_iter *iter = seq->private;
2429 
2430 	rcu_read_lock_bh();
2431 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2432 	iter->skip = *pos;
2433 
2434 	if (iter->tbl) {
2435 		ipv6_route_seq_setup_walk(iter, net);
2436 		return ipv6_route_seq_next(seq, NULL, pos);
2437 	} else {
2438 		return NULL;
2439 	}
2440 }
2441 
2442 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2443 {
2444 	struct fib6_walker *w = &iter->w;
2445 	return w->node && !(w->state == FWS_U && w->node == w->root);
2446 }
2447 
2448 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2449 	__releases(RCU_BH)
2450 {
2451 	struct net *net = seq_file_net(seq);
2452 	struct ipv6_route_iter *iter = seq->private;
2453 
2454 	if (ipv6_route_iter_active(iter))
2455 		fib6_walker_unlink(net, &iter->w);
2456 
2457 	rcu_read_unlock_bh();
2458 }
2459 
2460 const struct seq_operations ipv6_route_seq_ops = {
2461 	.start	= ipv6_route_seq_start,
2462 	.next	= ipv6_route_seq_next,
2463 	.stop	= ipv6_route_seq_stop,
2464 	.show	= ipv6_route_seq_show
2465 };
2466 #endif /* CONFIG_PROC_FS */
2467