xref: /linux/net/ipv6/ip6_fib.c (revision f6d08d9d8543c8ee494b307804b28e2750ffedb9)
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, bool with_fib6_nh)
147 {
148 	struct fib6_info *f6i;
149 	size_t sz = sizeof(*f6i);
150 
151 	if (with_fib6_nh)
152 		sz += sizeof(struct fib6_nh);
153 
154 	f6i = kzalloc(sz, gfp_flags);
155 	if (!f6i)
156 		return NULL;
157 
158 	/* fib6_siblings is a union with nh_list, so this initializes both */
159 	INIT_LIST_HEAD(&f6i->fib6_siblings);
160 	refcount_set(&f6i->fib6_ref, 1);
161 
162 	return f6i;
163 }
164 
165 void fib6_info_destroy_rcu(struct rcu_head *head)
166 {
167 	struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
168 
169 	WARN_ON(f6i->fib6_node);
170 
171 	if (f6i->nh)
172 		nexthop_put(f6i->nh);
173 	else
174 		fib6_nh_release(f6i->fib6_nh);
175 
176 	ip_fib_metrics_put(f6i->fib6_metrics);
177 	kfree(f6i);
178 }
179 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
180 
181 static struct fib6_node *node_alloc(struct net *net)
182 {
183 	struct fib6_node *fn;
184 
185 	fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
186 	if (fn)
187 		net->ipv6.rt6_stats->fib_nodes++;
188 
189 	return fn;
190 }
191 
192 static void node_free_immediate(struct net *net, struct fib6_node *fn)
193 {
194 	kmem_cache_free(fib6_node_kmem, fn);
195 	net->ipv6.rt6_stats->fib_nodes--;
196 }
197 
198 static void node_free_rcu(struct rcu_head *head)
199 {
200 	struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
201 
202 	kmem_cache_free(fib6_node_kmem, fn);
203 }
204 
205 static void node_free(struct net *net, struct fib6_node *fn)
206 {
207 	call_rcu(&fn->rcu, node_free_rcu);
208 	net->ipv6.rt6_stats->fib_nodes--;
209 }
210 
211 static void fib6_free_table(struct fib6_table *table)
212 {
213 	inetpeer_invalidate_tree(&table->tb6_peers);
214 	kfree(table);
215 }
216 
217 static void fib6_link_table(struct net *net, struct fib6_table *tb)
218 {
219 	unsigned int h;
220 
221 	/*
222 	 * Initialize table lock at a single place to give lockdep a key,
223 	 * tables aren't visible prior to being linked to the list.
224 	 */
225 	spin_lock_init(&tb->tb6_lock);
226 	h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
227 
228 	/*
229 	 * No protection necessary, this is the only list mutatation
230 	 * operation, tables never disappear once they exist.
231 	 */
232 	hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
233 }
234 
235 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
236 
237 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
238 {
239 	struct fib6_table *table;
240 
241 	table = kzalloc(sizeof(*table), GFP_ATOMIC);
242 	if (table) {
243 		table->tb6_id = id;
244 		rcu_assign_pointer(table->tb6_root.leaf,
245 				   net->ipv6.fib6_null_entry);
246 		table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
247 		inet_peer_base_init(&table->tb6_peers);
248 	}
249 
250 	return table;
251 }
252 
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
254 {
255 	struct fib6_table *tb;
256 
257 	if (id == 0)
258 		id = RT6_TABLE_MAIN;
259 	tb = fib6_get_table(net, id);
260 	if (tb)
261 		return tb;
262 
263 	tb = fib6_alloc_table(net, id);
264 	if (tb)
265 		fib6_link_table(net, tb);
266 
267 	return tb;
268 }
269 EXPORT_SYMBOL_GPL(fib6_new_table);
270 
271 struct fib6_table *fib6_get_table(struct net *net, u32 id)
272 {
273 	struct fib6_table *tb;
274 	struct hlist_head *head;
275 	unsigned int h;
276 
277 	if (id == 0)
278 		id = RT6_TABLE_MAIN;
279 	h = id & (FIB6_TABLE_HASHSZ - 1);
280 	rcu_read_lock();
281 	head = &net->ipv6.fib_table_hash[h];
282 	hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
283 		if (tb->tb6_id == id) {
284 			rcu_read_unlock();
285 			return tb;
286 		}
287 	}
288 	rcu_read_unlock();
289 
290 	return NULL;
291 }
292 EXPORT_SYMBOL_GPL(fib6_get_table);
293 
294 static void __net_init fib6_tables_init(struct net *net)
295 {
296 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
297 	fib6_link_table(net, net->ipv6.fib6_local_tbl);
298 }
299 #else
300 
301 struct fib6_table *fib6_new_table(struct net *net, u32 id)
302 {
303 	return fib6_get_table(net, id);
304 }
305 
306 struct fib6_table *fib6_get_table(struct net *net, u32 id)
307 {
308 	  return net->ipv6.fib6_main_tbl;
309 }
310 
311 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
312 				   const struct sk_buff *skb,
313 				   int flags, pol_lookup_t lookup)
314 {
315 	struct rt6_info *rt;
316 
317 	rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
318 	if (rt->dst.error == -EAGAIN) {
319 		ip6_rt_put_flags(rt, flags);
320 		rt = net->ipv6.ip6_null_entry;
321 		if (!(flags | RT6_LOOKUP_F_DST_NOREF))
322 			dst_hold(&rt->dst);
323 	}
324 
325 	return &rt->dst;
326 }
327 
328 /* called with rcu lock held; no reference taken on fib6_info */
329 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
330 		struct fib6_result *res, int flags)
331 {
332 	return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
333 				 res, flags);
334 }
335 
336 static void __net_init fib6_tables_init(struct net *net)
337 {
338 	fib6_link_table(net, net->ipv6.fib6_main_tbl);
339 }
340 
341 #endif
342 
343 unsigned int fib6_tables_seq_read(struct net *net)
344 {
345 	unsigned int h, fib_seq = 0;
346 
347 	rcu_read_lock();
348 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
349 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
350 		struct fib6_table *tb;
351 
352 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
353 			fib_seq += tb->fib_seq;
354 	}
355 	rcu_read_unlock();
356 
357 	return fib_seq;
358 }
359 
360 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
361 				    enum fib_event_type event_type,
362 				    struct fib6_info *rt)
363 {
364 	struct fib6_entry_notifier_info info = {
365 		.rt = rt,
366 	};
367 
368 	return call_fib6_notifier(nb, net, event_type, &info.info);
369 }
370 
371 int call_fib6_entry_notifiers(struct net *net,
372 			      enum fib_event_type event_type,
373 			      struct fib6_info *rt,
374 			      struct netlink_ext_ack *extack)
375 {
376 	struct fib6_entry_notifier_info info = {
377 		.info.extack = extack,
378 		.rt = rt,
379 	};
380 
381 	rt->fib6_table->fib_seq++;
382 	return call_fib6_notifiers(net, event_type, &info.info);
383 }
384 
385 int call_fib6_multipath_entry_notifiers(struct net *net,
386 					enum fib_event_type event_type,
387 					struct fib6_info *rt,
388 					unsigned int nsiblings,
389 					struct netlink_ext_ack *extack)
390 {
391 	struct fib6_entry_notifier_info info = {
392 		.info.extack = extack,
393 		.rt = rt,
394 		.nsiblings = nsiblings,
395 	};
396 
397 	rt->fib6_table->fib_seq++;
398 	return call_fib6_notifiers(net, event_type, &info.info);
399 }
400 
401 struct fib6_dump_arg {
402 	struct net *net;
403 	struct notifier_block *nb;
404 };
405 
406 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
407 {
408 	if (rt == arg->net->ipv6.fib6_null_entry)
409 		return;
410 	call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
411 }
412 
413 static int fib6_node_dump(struct fib6_walker *w)
414 {
415 	struct fib6_info *rt;
416 
417 	for_each_fib6_walker_rt(w)
418 		fib6_rt_dump(rt, w->args);
419 	w->leaf = NULL;
420 	return 0;
421 }
422 
423 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
424 			    struct fib6_walker *w)
425 {
426 	w->root = &tb->tb6_root;
427 	spin_lock_bh(&tb->tb6_lock);
428 	fib6_walk(net, w);
429 	spin_unlock_bh(&tb->tb6_lock);
430 }
431 
432 /* Called with rcu_read_lock() */
433 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
434 {
435 	struct fib6_dump_arg arg;
436 	struct fib6_walker *w;
437 	unsigned int h;
438 
439 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
440 	if (!w)
441 		return -ENOMEM;
442 
443 	w->func = fib6_node_dump;
444 	arg.net = net;
445 	arg.nb = nb;
446 	w->args = &arg;
447 
448 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
449 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
450 		struct fib6_table *tb;
451 
452 		hlist_for_each_entry_rcu(tb, head, tb6_hlist)
453 			fib6_table_dump(net, tb, w);
454 	}
455 
456 	kfree(w);
457 
458 	return 0;
459 }
460 
461 static int fib6_dump_node(struct fib6_walker *w)
462 {
463 	int res;
464 	struct fib6_info *rt;
465 
466 	for_each_fib6_walker_rt(w) {
467 		res = rt6_dump_route(rt, w->args);
468 		if (res < 0) {
469 			/* Frame is full, suspend walking */
470 			w->leaf = rt;
471 			return 1;
472 		}
473 
474 		/* Multipath routes are dumped in one route with the
475 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
476 		 * last sibling of this route (no need to dump the
477 		 * sibling routes again)
478 		 */
479 		if (rt->fib6_nsiblings)
480 			rt = list_last_entry(&rt->fib6_siblings,
481 					     struct fib6_info,
482 					     fib6_siblings);
483 	}
484 	w->leaf = NULL;
485 	return 0;
486 }
487 
488 static void fib6_dump_end(struct netlink_callback *cb)
489 {
490 	struct net *net = sock_net(cb->skb->sk);
491 	struct fib6_walker *w = (void *)cb->args[2];
492 
493 	if (w) {
494 		if (cb->args[4]) {
495 			cb->args[4] = 0;
496 			fib6_walker_unlink(net, w);
497 		}
498 		cb->args[2] = 0;
499 		kfree(w);
500 	}
501 	cb->done = (void *)cb->args[3];
502 	cb->args[1] = 3;
503 }
504 
505 static int fib6_dump_done(struct netlink_callback *cb)
506 {
507 	fib6_dump_end(cb);
508 	return cb->done ? cb->done(cb) : 0;
509 }
510 
511 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
512 			   struct netlink_callback *cb)
513 {
514 	struct net *net = sock_net(skb->sk);
515 	struct fib6_walker *w;
516 	int res;
517 
518 	w = (void *)cb->args[2];
519 	w->root = &table->tb6_root;
520 
521 	if (cb->args[4] == 0) {
522 		w->count = 0;
523 		w->skip = 0;
524 
525 		spin_lock_bh(&table->tb6_lock);
526 		res = fib6_walk(net, w);
527 		spin_unlock_bh(&table->tb6_lock);
528 		if (res > 0) {
529 			cb->args[4] = 1;
530 			cb->args[5] = w->root->fn_sernum;
531 		}
532 	} else {
533 		if (cb->args[5] != w->root->fn_sernum) {
534 			/* Begin at the root if the tree changed */
535 			cb->args[5] = w->root->fn_sernum;
536 			w->state = FWS_INIT;
537 			w->node = w->root;
538 			w->skip = w->count;
539 		} else
540 			w->skip = 0;
541 
542 		spin_lock_bh(&table->tb6_lock);
543 		res = fib6_walk_continue(w);
544 		spin_unlock_bh(&table->tb6_lock);
545 		if (res <= 0) {
546 			fib6_walker_unlink(net, w);
547 			cb->args[4] = 0;
548 		}
549 	}
550 
551 	return res;
552 }
553 
554 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
555 {
556 	const struct nlmsghdr *nlh = cb->nlh;
557 	struct net *net = sock_net(skb->sk);
558 	struct rt6_rtnl_dump_arg arg = {};
559 	unsigned int h, s_h;
560 	unsigned int e = 0, s_e;
561 	struct fib6_walker *w;
562 	struct fib6_table *tb;
563 	struct hlist_head *head;
564 	int res = 0;
565 
566 	if (cb->strict_check) {
567 		int err;
568 
569 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
570 		if (err < 0)
571 			return err;
572 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
573 		struct rtmsg *rtm = nlmsg_data(nlh);
574 
575 		arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
576 	}
577 
578 	/* fib entries are never clones */
579 	if (arg.filter.flags & RTM_F_CLONED)
580 		goto out;
581 
582 	w = (void *)cb->args[2];
583 	if (!w) {
584 		/* New dump:
585 		 *
586 		 * 1. hook callback destructor.
587 		 */
588 		cb->args[3] = (long)cb->done;
589 		cb->done = fib6_dump_done;
590 
591 		/*
592 		 * 2. allocate and initialize walker.
593 		 */
594 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
595 		if (!w)
596 			return -ENOMEM;
597 		w->func = fib6_dump_node;
598 		cb->args[2] = (long)w;
599 	}
600 
601 	arg.skb = skb;
602 	arg.cb = cb;
603 	arg.net = net;
604 	w->args = &arg;
605 
606 	if (arg.filter.table_id) {
607 		tb = fib6_get_table(net, arg.filter.table_id);
608 		if (!tb) {
609 			if (arg.filter.dump_all_families)
610 				goto out;
611 
612 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
613 			return -ENOENT;
614 		}
615 
616 		if (!cb->args[0]) {
617 			res = fib6_dump_table(tb, skb, cb);
618 			if (!res)
619 				cb->args[0] = 1;
620 		}
621 		goto out;
622 	}
623 
624 	s_h = cb->args[0];
625 	s_e = cb->args[1];
626 
627 	rcu_read_lock();
628 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
629 		e = 0;
630 		head = &net->ipv6.fib_table_hash[h];
631 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
632 			if (e < s_e)
633 				goto next;
634 			res = fib6_dump_table(tb, skb, cb);
635 			if (res != 0)
636 				goto out_unlock;
637 next:
638 			e++;
639 		}
640 	}
641 out_unlock:
642 	rcu_read_unlock();
643 	cb->args[1] = e;
644 	cb->args[0] = h;
645 out:
646 	res = res < 0 ? res : skb->len;
647 	if (res <= 0)
648 		fib6_dump_end(cb);
649 	return res;
650 }
651 
652 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
653 {
654 	if (!f6i)
655 		return;
656 
657 	if (f6i->fib6_metrics == &dst_default_metrics) {
658 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
659 
660 		if (!p)
661 			return;
662 
663 		refcount_set(&p->refcnt, 1);
664 		f6i->fib6_metrics = p;
665 	}
666 
667 	f6i->fib6_metrics->metrics[metric - 1] = val;
668 }
669 
670 /*
671  *	Routing Table
672  *
673  *	return the appropriate node for a routing tree "add" operation
674  *	by either creating and inserting or by returning an existing
675  *	node.
676  */
677 
678 static struct fib6_node *fib6_add_1(struct net *net,
679 				    struct fib6_table *table,
680 				    struct fib6_node *root,
681 				    struct in6_addr *addr, int plen,
682 				    int offset, int allow_create,
683 				    int replace_required,
684 				    struct netlink_ext_ack *extack)
685 {
686 	struct fib6_node *fn, *in, *ln;
687 	struct fib6_node *pn = NULL;
688 	struct rt6key *key;
689 	int	bit;
690 	__be32	dir = 0;
691 
692 	RT6_TRACE("fib6_add_1\n");
693 
694 	/* insert node in tree */
695 
696 	fn = root;
697 
698 	do {
699 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
700 					    lockdep_is_held(&table->tb6_lock));
701 		key = (struct rt6key *)((u8 *)leaf + offset);
702 
703 		/*
704 		 *	Prefix match
705 		 */
706 		if (plen < fn->fn_bit ||
707 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
708 			if (!allow_create) {
709 				if (replace_required) {
710 					NL_SET_ERR_MSG(extack,
711 						       "Can not replace route - no match found");
712 					pr_warn("Can't replace route, no match found\n");
713 					return ERR_PTR(-ENOENT);
714 				}
715 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
716 			}
717 			goto insert_above;
718 		}
719 
720 		/*
721 		 *	Exact match ?
722 		 */
723 
724 		if (plen == fn->fn_bit) {
725 			/* clean up an intermediate node */
726 			if (!(fn->fn_flags & RTN_RTINFO)) {
727 				RCU_INIT_POINTER(fn->leaf, NULL);
728 				fib6_info_release(leaf);
729 			/* remove null_entry in the root node */
730 			} else if (fn->fn_flags & RTN_TL_ROOT &&
731 				   rcu_access_pointer(fn->leaf) ==
732 				   net->ipv6.fib6_null_entry) {
733 				RCU_INIT_POINTER(fn->leaf, NULL);
734 			}
735 
736 			return fn;
737 		}
738 
739 		/*
740 		 *	We have more bits to go
741 		 */
742 
743 		/* Try to walk down on tree. */
744 		dir = addr_bit_set(addr, fn->fn_bit);
745 		pn = fn;
746 		fn = dir ?
747 		     rcu_dereference_protected(fn->right,
748 					lockdep_is_held(&table->tb6_lock)) :
749 		     rcu_dereference_protected(fn->left,
750 					lockdep_is_held(&table->tb6_lock));
751 	} while (fn);
752 
753 	if (!allow_create) {
754 		/* We should not create new node because
755 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
756 		 * I assume it is safe to require NLM_F_CREATE when
757 		 * REPLACE flag is used! Later we may want to remove the
758 		 * check for replace_required, because according
759 		 * to netlink specification, NLM_F_CREATE
760 		 * MUST be specified if new route is created.
761 		 * That would keep IPv6 consistent with IPv4
762 		 */
763 		if (replace_required) {
764 			NL_SET_ERR_MSG(extack,
765 				       "Can not replace route - no match found");
766 			pr_warn("Can't replace route, no match found\n");
767 			return ERR_PTR(-ENOENT);
768 		}
769 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
770 	}
771 	/*
772 	 *	We walked to the bottom of tree.
773 	 *	Create new leaf node without children.
774 	 */
775 
776 	ln = node_alloc(net);
777 
778 	if (!ln)
779 		return ERR_PTR(-ENOMEM);
780 	ln->fn_bit = plen;
781 	RCU_INIT_POINTER(ln->parent, pn);
782 
783 	if (dir)
784 		rcu_assign_pointer(pn->right, ln);
785 	else
786 		rcu_assign_pointer(pn->left, ln);
787 
788 	return ln;
789 
790 
791 insert_above:
792 	/*
793 	 * split since we don't have a common prefix anymore or
794 	 * we have a less significant route.
795 	 * we've to insert an intermediate node on the list
796 	 * this new node will point to the one we need to create
797 	 * and the current
798 	 */
799 
800 	pn = rcu_dereference_protected(fn->parent,
801 				       lockdep_is_held(&table->tb6_lock));
802 
803 	/* find 1st bit in difference between the 2 addrs.
804 
805 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
806 	   but if it is >= plen, the value is ignored in any case.
807 	 */
808 
809 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
810 
811 	/*
812 	 *		(intermediate)[in]
813 	 *	          /	   \
814 	 *	(new leaf node)[ln] (old node)[fn]
815 	 */
816 	if (plen > bit) {
817 		in = node_alloc(net);
818 		ln = node_alloc(net);
819 
820 		if (!in || !ln) {
821 			if (in)
822 				node_free_immediate(net, in);
823 			if (ln)
824 				node_free_immediate(net, ln);
825 			return ERR_PTR(-ENOMEM);
826 		}
827 
828 		/*
829 		 * new intermediate node.
830 		 * RTN_RTINFO will
831 		 * be off since that an address that chooses one of
832 		 * the branches would not match less specific routes
833 		 * in the other branch
834 		 */
835 
836 		in->fn_bit = bit;
837 
838 		RCU_INIT_POINTER(in->parent, pn);
839 		in->leaf = fn->leaf;
840 		fib6_info_hold(rcu_dereference_protected(in->leaf,
841 				lockdep_is_held(&table->tb6_lock)));
842 
843 		/* update parent pointer */
844 		if (dir)
845 			rcu_assign_pointer(pn->right, in);
846 		else
847 			rcu_assign_pointer(pn->left, in);
848 
849 		ln->fn_bit = plen;
850 
851 		RCU_INIT_POINTER(ln->parent, in);
852 		rcu_assign_pointer(fn->parent, in);
853 
854 		if (addr_bit_set(addr, bit)) {
855 			rcu_assign_pointer(in->right, ln);
856 			rcu_assign_pointer(in->left, fn);
857 		} else {
858 			rcu_assign_pointer(in->left, ln);
859 			rcu_assign_pointer(in->right, fn);
860 		}
861 	} else { /* plen <= bit */
862 
863 		/*
864 		 *		(new leaf node)[ln]
865 		 *	          /	   \
866 		 *	     (old node)[fn] NULL
867 		 */
868 
869 		ln = node_alloc(net);
870 
871 		if (!ln)
872 			return ERR_PTR(-ENOMEM);
873 
874 		ln->fn_bit = plen;
875 
876 		RCU_INIT_POINTER(ln->parent, pn);
877 
878 		if (addr_bit_set(&key->addr, plen))
879 			RCU_INIT_POINTER(ln->right, fn);
880 		else
881 			RCU_INIT_POINTER(ln->left, fn);
882 
883 		rcu_assign_pointer(fn->parent, ln);
884 
885 		if (dir)
886 			rcu_assign_pointer(pn->right, ln);
887 		else
888 			rcu_assign_pointer(pn->left, ln);
889 	}
890 	return ln;
891 }
892 
893 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
894 				  const struct fib6_info *match,
895 				  const struct fib6_table *table)
896 {
897 	int cpu;
898 
899 	if (!fib6_nh->rt6i_pcpu)
900 		return;
901 
902 	/* release the reference to this fib entry from
903 	 * all of its cached pcpu routes
904 	 */
905 	for_each_possible_cpu(cpu) {
906 		struct rt6_info **ppcpu_rt;
907 		struct rt6_info *pcpu_rt;
908 
909 		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
910 		pcpu_rt = *ppcpu_rt;
911 
912 		/* only dropping the 'from' reference if the cached route
913 		 * is using 'match'. The cached pcpu_rt->from only changes
914 		 * from a fib6_info to NULL (ip6_dst_destroy); it can never
915 		 * change from one fib6_info reference to another
916 		 */
917 		if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
918 			struct fib6_info *from;
919 
920 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
921 			fib6_info_release(from);
922 		}
923 	}
924 }
925 
926 struct fib6_nh_pcpu_arg {
927 	struct fib6_info	*from;
928 	const struct fib6_table *table;
929 };
930 
931 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
932 {
933 	struct fib6_nh_pcpu_arg *arg = _arg;
934 
935 	__fib6_drop_pcpu_from(nh, arg->from, arg->table);
936 	return 0;
937 }
938 
939 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
940 				const struct fib6_table *table)
941 {
942 	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
943 	 * while we are cleaning them here.
944 	 */
945 	f6i->fib6_destroying = 1;
946 	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
947 
948 	if (f6i->nh) {
949 		struct fib6_nh_pcpu_arg arg = {
950 			.from = f6i,
951 			.table = table
952 		};
953 
954 		nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
955 					 &arg);
956 	} else {
957 		struct fib6_nh *fib6_nh;
958 
959 		fib6_nh = f6i->fib6_nh;
960 		__fib6_drop_pcpu_from(fib6_nh, f6i, table);
961 	}
962 }
963 
964 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
965 			  struct net *net)
966 {
967 	struct fib6_table *table = rt->fib6_table;
968 
969 	fib6_drop_pcpu_from(rt, table);
970 
971 	if (rt->nh && !list_empty(&rt->nh_list))
972 		list_del_init(&rt->nh_list);
973 
974 	if (refcount_read(&rt->fib6_ref) != 1) {
975 		/* This route is used as dummy address holder in some split
976 		 * nodes. It is not leaked, but it still holds other resources,
977 		 * which must be released in time. So, scan ascendant nodes
978 		 * and replace dummy references to this route with references
979 		 * to still alive ones.
980 		 */
981 		while (fn) {
982 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
983 					    lockdep_is_held(&table->tb6_lock));
984 			struct fib6_info *new_leaf;
985 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
986 				new_leaf = fib6_find_prefix(net, table, fn);
987 				fib6_info_hold(new_leaf);
988 
989 				rcu_assign_pointer(fn->leaf, new_leaf);
990 				fib6_info_release(rt);
991 			}
992 			fn = rcu_dereference_protected(fn->parent,
993 				    lockdep_is_held(&table->tb6_lock));
994 		}
995 	}
996 }
997 
998 /*
999  *	Insert routing information in a node.
1000  */
1001 
1002 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1003 			    struct nl_info *info,
1004 			    struct netlink_ext_ack *extack)
1005 {
1006 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1007 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1008 	struct fib6_info *iter = NULL;
1009 	struct fib6_info __rcu **ins;
1010 	struct fib6_info __rcu **fallback_ins = NULL;
1011 	int replace = (info->nlh &&
1012 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1013 	int add = (!info->nlh ||
1014 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
1015 	int found = 0;
1016 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1017 	u16 nlflags = NLM_F_EXCL;
1018 	int err;
1019 
1020 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1021 		nlflags |= NLM_F_APPEND;
1022 
1023 	ins = &fn->leaf;
1024 
1025 	for (iter = leaf; iter;
1026 	     iter = rcu_dereference_protected(iter->fib6_next,
1027 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1028 		/*
1029 		 *	Search for duplicates
1030 		 */
1031 
1032 		if (iter->fib6_metric == rt->fib6_metric) {
1033 			/*
1034 			 *	Same priority level
1035 			 */
1036 			if (info->nlh &&
1037 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1038 				return -EEXIST;
1039 
1040 			nlflags &= ~NLM_F_EXCL;
1041 			if (replace) {
1042 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1043 					found++;
1044 					break;
1045 				}
1046 				if (rt_can_ecmp)
1047 					fallback_ins = fallback_ins ?: ins;
1048 				goto next_iter;
1049 			}
1050 
1051 			if (rt6_duplicate_nexthop(iter, rt)) {
1052 				if (rt->fib6_nsiblings)
1053 					rt->fib6_nsiblings = 0;
1054 				if (!(iter->fib6_flags & RTF_EXPIRES))
1055 					return -EEXIST;
1056 				if (!(rt->fib6_flags & RTF_EXPIRES))
1057 					fib6_clean_expires(iter);
1058 				else
1059 					fib6_set_expires(iter, rt->expires);
1060 
1061 				if (rt->fib6_pmtu)
1062 					fib6_metric_set(iter, RTAX_MTU,
1063 							rt->fib6_pmtu);
1064 				return -EEXIST;
1065 			}
1066 			/* If we have the same destination and the same metric,
1067 			 * but not the same gateway, then the route we try to
1068 			 * add is sibling to this route, increment our counter
1069 			 * of siblings, and later we will add our route to the
1070 			 * list.
1071 			 * Only static routes (which don't have flag
1072 			 * RTF_EXPIRES) are used for ECMPv6.
1073 			 *
1074 			 * To avoid long list, we only had siblings if the
1075 			 * route have a gateway.
1076 			 */
1077 			if (rt_can_ecmp &&
1078 			    rt6_qualify_for_ecmp(iter))
1079 				rt->fib6_nsiblings++;
1080 		}
1081 
1082 		if (iter->fib6_metric > rt->fib6_metric)
1083 			break;
1084 
1085 next_iter:
1086 		ins = &iter->fib6_next;
1087 	}
1088 
1089 	if (fallback_ins && !found) {
1090 		/* No ECMP-able route found, replace first non-ECMP one */
1091 		ins = fallback_ins;
1092 		iter = rcu_dereference_protected(*ins,
1093 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1094 		found++;
1095 	}
1096 
1097 	/* Reset round-robin state, if necessary */
1098 	if (ins == &fn->leaf)
1099 		fn->rr_ptr = NULL;
1100 
1101 	/* Link this route to others same route. */
1102 	if (rt->fib6_nsiblings) {
1103 		unsigned int fib6_nsiblings;
1104 		struct fib6_info *sibling, *temp_sibling;
1105 
1106 		/* Find the first route that have the same metric */
1107 		sibling = leaf;
1108 		while (sibling) {
1109 			if (sibling->fib6_metric == rt->fib6_metric &&
1110 			    rt6_qualify_for_ecmp(sibling)) {
1111 				list_add_tail(&rt->fib6_siblings,
1112 					      &sibling->fib6_siblings);
1113 				break;
1114 			}
1115 			sibling = rcu_dereference_protected(sibling->fib6_next,
1116 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1117 		}
1118 		/* For each sibling in the list, increment the counter of
1119 		 * siblings. BUG() if counters does not match, list of siblings
1120 		 * is broken!
1121 		 */
1122 		fib6_nsiblings = 0;
1123 		list_for_each_entry_safe(sibling, temp_sibling,
1124 					 &rt->fib6_siblings, fib6_siblings) {
1125 			sibling->fib6_nsiblings++;
1126 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1127 			fib6_nsiblings++;
1128 		}
1129 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1130 		rt6_multipath_rebalance(temp_sibling);
1131 	}
1132 
1133 	/*
1134 	 *	insert node
1135 	 */
1136 	if (!replace) {
1137 		if (!add)
1138 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1139 
1140 add:
1141 		nlflags |= NLM_F_CREATE;
1142 
1143 		if (!info->skip_notify_kernel) {
1144 			err = call_fib6_entry_notifiers(info->nl_net,
1145 							FIB_EVENT_ENTRY_ADD,
1146 							rt, extack);
1147 			if (err)
1148 				return err;
1149 		}
1150 
1151 		rcu_assign_pointer(rt->fib6_next, iter);
1152 		fib6_info_hold(rt);
1153 		rcu_assign_pointer(rt->fib6_node, fn);
1154 		rcu_assign_pointer(*ins, rt);
1155 		if (!info->skip_notify)
1156 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1157 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1158 
1159 		if (!(fn->fn_flags & RTN_RTINFO)) {
1160 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1161 			fn->fn_flags |= RTN_RTINFO;
1162 		}
1163 
1164 	} else {
1165 		int nsiblings;
1166 
1167 		if (!found) {
1168 			if (add)
1169 				goto add;
1170 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1171 			return -ENOENT;
1172 		}
1173 
1174 		if (!info->skip_notify_kernel) {
1175 			err = call_fib6_entry_notifiers(info->nl_net,
1176 							FIB_EVENT_ENTRY_REPLACE,
1177 							rt, extack);
1178 			if (err)
1179 				return err;
1180 		}
1181 
1182 		fib6_info_hold(rt);
1183 		rcu_assign_pointer(rt->fib6_node, fn);
1184 		rt->fib6_next = iter->fib6_next;
1185 		rcu_assign_pointer(*ins, rt);
1186 		if (!info->skip_notify)
1187 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1188 		if (!(fn->fn_flags & RTN_RTINFO)) {
1189 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1190 			fn->fn_flags |= RTN_RTINFO;
1191 		}
1192 		nsiblings = iter->fib6_nsiblings;
1193 		iter->fib6_node = NULL;
1194 		fib6_purge_rt(iter, fn, info->nl_net);
1195 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1196 			fn->rr_ptr = NULL;
1197 		fib6_info_release(iter);
1198 
1199 		if (nsiblings) {
1200 			/* Replacing an ECMP route, remove all siblings */
1201 			ins = &rt->fib6_next;
1202 			iter = rcu_dereference_protected(*ins,
1203 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1204 			while (iter) {
1205 				if (iter->fib6_metric > rt->fib6_metric)
1206 					break;
1207 				if (rt6_qualify_for_ecmp(iter)) {
1208 					*ins = iter->fib6_next;
1209 					iter->fib6_node = NULL;
1210 					fib6_purge_rt(iter, fn, info->nl_net);
1211 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1212 						fn->rr_ptr = NULL;
1213 					fib6_info_release(iter);
1214 					nsiblings--;
1215 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1216 				} else {
1217 					ins = &iter->fib6_next;
1218 				}
1219 				iter = rcu_dereference_protected(*ins,
1220 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1221 			}
1222 			WARN_ON(nsiblings != 0);
1223 		}
1224 	}
1225 
1226 	return 0;
1227 }
1228 
1229 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1230 {
1231 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1232 	    (rt->fib6_flags & RTF_EXPIRES))
1233 		mod_timer(&net->ipv6.ip6_fib_timer,
1234 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1235 }
1236 
1237 void fib6_force_start_gc(struct net *net)
1238 {
1239 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1240 		mod_timer(&net->ipv6.ip6_fib_timer,
1241 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1242 }
1243 
1244 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1245 					   int sernum)
1246 {
1247 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1248 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1249 
1250 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1251 	smp_wmb();
1252 	while (fn) {
1253 		fn->fn_sernum = sernum;
1254 		fn = rcu_dereference_protected(fn->parent,
1255 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1256 	}
1257 }
1258 
1259 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1260 {
1261 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1262 }
1263 
1264 /* allow ipv4 to update sernum via ipv6_stub */
1265 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1266 {
1267 	spin_lock_bh(&f6i->fib6_table->tb6_lock);
1268 	fib6_update_sernum_upto_root(net, f6i);
1269 	spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1270 }
1271 
1272 /*
1273  *	Add routing information to the routing tree.
1274  *	<destination addr>/<source addr>
1275  *	with source addr info in sub-trees
1276  *	Need to own table->tb6_lock
1277  */
1278 
1279 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1280 	     struct nl_info *info, struct netlink_ext_ack *extack)
1281 {
1282 	struct fib6_table *table = rt->fib6_table;
1283 	struct fib6_node *fn, *pn = NULL;
1284 	int err = -ENOMEM;
1285 	int allow_create = 1;
1286 	int replace_required = 0;
1287 	int sernum = fib6_new_sernum(info->nl_net);
1288 
1289 	if (info->nlh) {
1290 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1291 			allow_create = 0;
1292 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1293 			replace_required = 1;
1294 	}
1295 	if (!allow_create && !replace_required)
1296 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1297 
1298 	fn = fib6_add_1(info->nl_net, table, root,
1299 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1300 			offsetof(struct fib6_info, fib6_dst), allow_create,
1301 			replace_required, extack);
1302 	if (IS_ERR(fn)) {
1303 		err = PTR_ERR(fn);
1304 		fn = NULL;
1305 		goto out;
1306 	}
1307 
1308 	pn = fn;
1309 
1310 #ifdef CONFIG_IPV6_SUBTREES
1311 	if (rt->fib6_src.plen) {
1312 		struct fib6_node *sn;
1313 
1314 		if (!rcu_access_pointer(fn->subtree)) {
1315 			struct fib6_node *sfn;
1316 
1317 			/*
1318 			 * Create subtree.
1319 			 *
1320 			 *		fn[main tree]
1321 			 *		|
1322 			 *		sfn[subtree root]
1323 			 *		   \
1324 			 *		    sn[new leaf node]
1325 			 */
1326 
1327 			/* Create subtree root node */
1328 			sfn = node_alloc(info->nl_net);
1329 			if (!sfn)
1330 				goto failure;
1331 
1332 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1333 			rcu_assign_pointer(sfn->leaf,
1334 					   info->nl_net->ipv6.fib6_null_entry);
1335 			sfn->fn_flags = RTN_ROOT;
1336 
1337 			/* Now add the first leaf node to new subtree */
1338 
1339 			sn = fib6_add_1(info->nl_net, table, sfn,
1340 					&rt->fib6_src.addr, rt->fib6_src.plen,
1341 					offsetof(struct fib6_info, fib6_src),
1342 					allow_create, replace_required, extack);
1343 
1344 			if (IS_ERR(sn)) {
1345 				/* If it is failed, discard just allocated
1346 				   root, and then (in failure) stale node
1347 				   in main tree.
1348 				 */
1349 				node_free_immediate(info->nl_net, sfn);
1350 				err = PTR_ERR(sn);
1351 				goto failure;
1352 			}
1353 
1354 			/* Now link new subtree to main tree */
1355 			rcu_assign_pointer(sfn->parent, fn);
1356 			rcu_assign_pointer(fn->subtree, sfn);
1357 		} else {
1358 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1359 					&rt->fib6_src.addr, rt->fib6_src.plen,
1360 					offsetof(struct fib6_info, fib6_src),
1361 					allow_create, replace_required, extack);
1362 
1363 			if (IS_ERR(sn)) {
1364 				err = PTR_ERR(sn);
1365 				goto failure;
1366 			}
1367 		}
1368 
1369 		if (!rcu_access_pointer(fn->leaf)) {
1370 			if (fn->fn_flags & RTN_TL_ROOT) {
1371 				/* put back null_entry for root node */
1372 				rcu_assign_pointer(fn->leaf,
1373 					    info->nl_net->ipv6.fib6_null_entry);
1374 			} else {
1375 				fib6_info_hold(rt);
1376 				rcu_assign_pointer(fn->leaf, rt);
1377 			}
1378 		}
1379 		fn = sn;
1380 	}
1381 #endif
1382 
1383 	err = fib6_add_rt2node(fn, rt, info, extack);
1384 	if (!err) {
1385 		if (rt->nh)
1386 			list_add(&rt->nh_list, &rt->nh->f6i_list);
1387 		__fib6_update_sernum_upto_root(rt, sernum);
1388 		fib6_start_gc(info->nl_net, rt);
1389 	}
1390 
1391 out:
1392 	if (err) {
1393 #ifdef CONFIG_IPV6_SUBTREES
1394 		/*
1395 		 * If fib6_add_1 has cleared the old leaf pointer in the
1396 		 * super-tree leaf node we have to find a new one for it.
1397 		 */
1398 		if (pn != fn) {
1399 			struct fib6_info *pn_leaf =
1400 				rcu_dereference_protected(pn->leaf,
1401 				    lockdep_is_held(&table->tb6_lock));
1402 			if (pn_leaf == rt) {
1403 				pn_leaf = NULL;
1404 				RCU_INIT_POINTER(pn->leaf, NULL);
1405 				fib6_info_release(rt);
1406 			}
1407 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1408 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1409 							   pn);
1410 #if RT6_DEBUG >= 2
1411 				if (!pn_leaf) {
1412 					WARN_ON(!pn_leaf);
1413 					pn_leaf =
1414 					    info->nl_net->ipv6.fib6_null_entry;
1415 				}
1416 #endif
1417 				fib6_info_hold(pn_leaf);
1418 				rcu_assign_pointer(pn->leaf, pn_leaf);
1419 			}
1420 		}
1421 #endif
1422 		goto failure;
1423 	}
1424 	return err;
1425 
1426 failure:
1427 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1428 	 * 1. fn is an intermediate node and we failed to add the new
1429 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1430 	 * failure case.
1431 	 * 2. fn is the root node in the table and we fail to add the first
1432 	 * default route to it.
1433 	 */
1434 	if (fn &&
1435 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1436 	     (fn->fn_flags & RTN_TL_ROOT &&
1437 	      !rcu_access_pointer(fn->leaf))))
1438 		fib6_repair_tree(info->nl_net, table, fn);
1439 	return err;
1440 }
1441 
1442 /*
1443  *	Routing tree lookup
1444  *
1445  */
1446 
1447 struct lookup_args {
1448 	int			offset;		/* key offset on fib6_info */
1449 	const struct in6_addr	*addr;		/* search key			*/
1450 };
1451 
1452 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1453 					    struct lookup_args *args)
1454 {
1455 	struct fib6_node *fn;
1456 	__be32 dir;
1457 
1458 	if (unlikely(args->offset == 0))
1459 		return NULL;
1460 
1461 	/*
1462 	 *	Descend on a tree
1463 	 */
1464 
1465 	fn = root;
1466 
1467 	for (;;) {
1468 		struct fib6_node *next;
1469 
1470 		dir = addr_bit_set(args->addr, fn->fn_bit);
1471 
1472 		next = dir ? rcu_dereference(fn->right) :
1473 			     rcu_dereference(fn->left);
1474 
1475 		if (next) {
1476 			fn = next;
1477 			continue;
1478 		}
1479 		break;
1480 	}
1481 
1482 	while (fn) {
1483 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1484 
1485 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1486 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1487 			struct rt6key *key;
1488 
1489 			if (!leaf)
1490 				goto backtrack;
1491 
1492 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1493 
1494 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1495 #ifdef CONFIG_IPV6_SUBTREES
1496 				if (subtree) {
1497 					struct fib6_node *sfn;
1498 					sfn = fib6_node_lookup_1(subtree,
1499 								 args + 1);
1500 					if (!sfn)
1501 						goto backtrack;
1502 					fn = sfn;
1503 				}
1504 #endif
1505 				if (fn->fn_flags & RTN_RTINFO)
1506 					return fn;
1507 			}
1508 		}
1509 backtrack:
1510 		if (fn->fn_flags & RTN_ROOT)
1511 			break;
1512 
1513 		fn = rcu_dereference(fn->parent);
1514 	}
1515 
1516 	return NULL;
1517 }
1518 
1519 /* called with rcu_read_lock() held
1520  */
1521 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1522 				   const struct in6_addr *daddr,
1523 				   const struct in6_addr *saddr)
1524 {
1525 	struct fib6_node *fn;
1526 	struct lookup_args args[] = {
1527 		{
1528 			.offset = offsetof(struct fib6_info, fib6_dst),
1529 			.addr = daddr,
1530 		},
1531 #ifdef CONFIG_IPV6_SUBTREES
1532 		{
1533 			.offset = offsetof(struct fib6_info, fib6_src),
1534 			.addr = saddr,
1535 		},
1536 #endif
1537 		{
1538 			.offset = 0,	/* sentinel */
1539 		}
1540 	};
1541 
1542 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1543 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1544 		fn = root;
1545 
1546 	return fn;
1547 }
1548 
1549 /*
1550  *	Get node with specified destination prefix (and source prefix,
1551  *	if subtrees are used)
1552  *	exact_match == true means we try to find fn with exact match of
1553  *	the passed in prefix addr
1554  *	exact_match == false means we try to find fn with longest prefix
1555  *	match of the passed in prefix addr. This is useful for finding fn
1556  *	for cached route as it will be stored in the exception table under
1557  *	the node with longest prefix length.
1558  */
1559 
1560 
1561 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1562 				       const struct in6_addr *addr,
1563 				       int plen, int offset,
1564 				       bool exact_match)
1565 {
1566 	struct fib6_node *fn, *prev = NULL;
1567 
1568 	for (fn = root; fn ; ) {
1569 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1570 		struct rt6key *key;
1571 
1572 		/* This node is being deleted */
1573 		if (!leaf) {
1574 			if (plen <= fn->fn_bit)
1575 				goto out;
1576 			else
1577 				goto next;
1578 		}
1579 
1580 		key = (struct rt6key *)((u8 *)leaf + offset);
1581 
1582 		/*
1583 		 *	Prefix match
1584 		 */
1585 		if (plen < fn->fn_bit ||
1586 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1587 			goto out;
1588 
1589 		if (plen == fn->fn_bit)
1590 			return fn;
1591 
1592 		prev = fn;
1593 
1594 next:
1595 		/*
1596 		 *	We have more bits to go
1597 		 */
1598 		if (addr_bit_set(addr, fn->fn_bit))
1599 			fn = rcu_dereference(fn->right);
1600 		else
1601 			fn = rcu_dereference(fn->left);
1602 	}
1603 out:
1604 	if (exact_match)
1605 		return NULL;
1606 	else
1607 		return prev;
1608 }
1609 
1610 struct fib6_node *fib6_locate(struct fib6_node *root,
1611 			      const struct in6_addr *daddr, int dst_len,
1612 			      const struct in6_addr *saddr, int src_len,
1613 			      bool exact_match)
1614 {
1615 	struct fib6_node *fn;
1616 
1617 	fn = fib6_locate_1(root, daddr, dst_len,
1618 			   offsetof(struct fib6_info, fib6_dst),
1619 			   exact_match);
1620 
1621 #ifdef CONFIG_IPV6_SUBTREES
1622 	if (src_len) {
1623 		WARN_ON(saddr == NULL);
1624 		if (fn) {
1625 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1626 
1627 			if (subtree) {
1628 				fn = fib6_locate_1(subtree, saddr, src_len,
1629 					   offsetof(struct fib6_info, fib6_src),
1630 					   exact_match);
1631 			}
1632 		}
1633 	}
1634 #endif
1635 
1636 	if (fn && fn->fn_flags & RTN_RTINFO)
1637 		return fn;
1638 
1639 	return NULL;
1640 }
1641 
1642 
1643 /*
1644  *	Deletion
1645  *
1646  */
1647 
1648 static struct fib6_info *fib6_find_prefix(struct net *net,
1649 					 struct fib6_table *table,
1650 					 struct fib6_node *fn)
1651 {
1652 	struct fib6_node *child_left, *child_right;
1653 
1654 	if (fn->fn_flags & RTN_ROOT)
1655 		return net->ipv6.fib6_null_entry;
1656 
1657 	while (fn) {
1658 		child_left = rcu_dereference_protected(fn->left,
1659 				    lockdep_is_held(&table->tb6_lock));
1660 		child_right = rcu_dereference_protected(fn->right,
1661 				    lockdep_is_held(&table->tb6_lock));
1662 		if (child_left)
1663 			return rcu_dereference_protected(child_left->leaf,
1664 					lockdep_is_held(&table->tb6_lock));
1665 		if (child_right)
1666 			return rcu_dereference_protected(child_right->leaf,
1667 					lockdep_is_held(&table->tb6_lock));
1668 
1669 		fn = FIB6_SUBTREE(fn);
1670 	}
1671 	return NULL;
1672 }
1673 
1674 /*
1675  *	Called to trim the tree of intermediate nodes when possible. "fn"
1676  *	is the node we want to try and remove.
1677  *	Need to own table->tb6_lock
1678  */
1679 
1680 static struct fib6_node *fib6_repair_tree(struct net *net,
1681 					  struct fib6_table *table,
1682 					  struct fib6_node *fn)
1683 {
1684 	int children;
1685 	int nstate;
1686 	struct fib6_node *child;
1687 	struct fib6_walker *w;
1688 	int iter = 0;
1689 
1690 	/* Set fn->leaf to null_entry for root node. */
1691 	if (fn->fn_flags & RTN_TL_ROOT) {
1692 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1693 		return fn;
1694 	}
1695 
1696 	for (;;) {
1697 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1698 					    lockdep_is_held(&table->tb6_lock));
1699 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1700 					    lockdep_is_held(&table->tb6_lock));
1701 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1702 					    lockdep_is_held(&table->tb6_lock));
1703 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1704 					    lockdep_is_held(&table->tb6_lock));
1705 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1706 					    lockdep_is_held(&table->tb6_lock));
1707 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1708 					    lockdep_is_held(&table->tb6_lock));
1709 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1710 					    lockdep_is_held(&table->tb6_lock));
1711 		struct fib6_info *new_fn_leaf;
1712 
1713 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1714 		iter++;
1715 
1716 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1717 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1718 		WARN_ON(fn_leaf);
1719 
1720 		children = 0;
1721 		child = NULL;
1722 		if (fn_r)
1723 			child = fn_r, children |= 1;
1724 		if (fn_l)
1725 			child = fn_l, children |= 2;
1726 
1727 		if (children == 3 || FIB6_SUBTREE(fn)
1728 #ifdef CONFIG_IPV6_SUBTREES
1729 		    /* Subtree root (i.e. fn) may have one child */
1730 		    || (children && fn->fn_flags & RTN_ROOT)
1731 #endif
1732 		    ) {
1733 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1734 #if RT6_DEBUG >= 2
1735 			if (!new_fn_leaf) {
1736 				WARN_ON(!new_fn_leaf);
1737 				new_fn_leaf = net->ipv6.fib6_null_entry;
1738 			}
1739 #endif
1740 			fib6_info_hold(new_fn_leaf);
1741 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1742 			return pn;
1743 		}
1744 
1745 #ifdef CONFIG_IPV6_SUBTREES
1746 		if (FIB6_SUBTREE(pn) == fn) {
1747 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1748 			RCU_INIT_POINTER(pn->subtree, NULL);
1749 			nstate = FWS_L;
1750 		} else {
1751 			WARN_ON(fn->fn_flags & RTN_ROOT);
1752 #endif
1753 			if (pn_r == fn)
1754 				rcu_assign_pointer(pn->right, child);
1755 			else if (pn_l == fn)
1756 				rcu_assign_pointer(pn->left, child);
1757 #if RT6_DEBUG >= 2
1758 			else
1759 				WARN_ON(1);
1760 #endif
1761 			if (child)
1762 				rcu_assign_pointer(child->parent, pn);
1763 			nstate = FWS_R;
1764 #ifdef CONFIG_IPV6_SUBTREES
1765 		}
1766 #endif
1767 
1768 		read_lock(&net->ipv6.fib6_walker_lock);
1769 		FOR_WALKERS(net, w) {
1770 			if (!child) {
1771 				if (w->node == fn) {
1772 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1773 					w->node = pn;
1774 					w->state = nstate;
1775 				}
1776 			} else {
1777 				if (w->node == fn) {
1778 					w->node = child;
1779 					if (children&2) {
1780 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1781 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1782 					} else {
1783 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1784 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1785 					}
1786 				}
1787 			}
1788 		}
1789 		read_unlock(&net->ipv6.fib6_walker_lock);
1790 
1791 		node_free(net, fn);
1792 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1793 			return pn;
1794 
1795 		RCU_INIT_POINTER(pn->leaf, NULL);
1796 		fib6_info_release(pn_leaf);
1797 		fn = pn;
1798 	}
1799 }
1800 
1801 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1802 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1803 {
1804 	struct fib6_walker *w;
1805 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1806 				    lockdep_is_held(&table->tb6_lock));
1807 	struct net *net = info->nl_net;
1808 
1809 	RT6_TRACE("fib6_del_route\n");
1810 
1811 	/* Unlink it */
1812 	*rtp = rt->fib6_next;
1813 	rt->fib6_node = NULL;
1814 	net->ipv6.rt6_stats->fib_rt_entries--;
1815 	net->ipv6.rt6_stats->fib_discarded_routes++;
1816 
1817 	/* Flush all cached dst in exception table */
1818 	rt6_flush_exceptions(rt);
1819 
1820 	/* Reset round-robin state, if necessary */
1821 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1822 		fn->rr_ptr = NULL;
1823 
1824 	/* Remove this entry from other siblings */
1825 	if (rt->fib6_nsiblings) {
1826 		struct fib6_info *sibling, *next_sibling;
1827 
1828 		list_for_each_entry_safe(sibling, next_sibling,
1829 					 &rt->fib6_siblings, fib6_siblings)
1830 			sibling->fib6_nsiblings--;
1831 		rt->fib6_nsiblings = 0;
1832 		list_del_init(&rt->fib6_siblings);
1833 		rt6_multipath_rebalance(next_sibling);
1834 	}
1835 
1836 	/* Adjust walkers */
1837 	read_lock(&net->ipv6.fib6_walker_lock);
1838 	FOR_WALKERS(net, w) {
1839 		if (w->state == FWS_C && w->leaf == rt) {
1840 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1841 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1842 					    lockdep_is_held(&table->tb6_lock));
1843 			if (!w->leaf)
1844 				w->state = FWS_U;
1845 		}
1846 	}
1847 	read_unlock(&net->ipv6.fib6_walker_lock);
1848 
1849 	/* If it was last route, call fib6_repair_tree() to:
1850 	 * 1. For root node, put back null_entry as how the table was created.
1851 	 * 2. For other nodes, expunge its radix tree node.
1852 	 */
1853 	if (!rcu_access_pointer(fn->leaf)) {
1854 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1855 			fn->fn_flags &= ~RTN_RTINFO;
1856 			net->ipv6.rt6_stats->fib_route_nodes--;
1857 		}
1858 		fn = fib6_repair_tree(net, table, fn);
1859 	}
1860 
1861 	fib6_purge_rt(rt, fn, net);
1862 
1863 	if (!info->skip_notify_kernel)
1864 		call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1865 	if (!info->skip_notify)
1866 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1867 
1868 	fib6_info_release(rt);
1869 }
1870 
1871 /* Need to own table->tb6_lock */
1872 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1873 {
1874 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1875 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1876 	struct fib6_table *table = rt->fib6_table;
1877 	struct net *net = info->nl_net;
1878 	struct fib6_info __rcu **rtp;
1879 	struct fib6_info __rcu **rtp_next;
1880 
1881 	if (!fn || rt == net->ipv6.fib6_null_entry)
1882 		return -ENOENT;
1883 
1884 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1885 
1886 	/*
1887 	 *	Walk the leaf entries looking for ourself
1888 	 */
1889 
1890 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1891 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1892 					lockdep_is_held(&table->tb6_lock));
1893 		if (rt == cur) {
1894 			fib6_del_route(table, fn, rtp, info);
1895 			return 0;
1896 		}
1897 		rtp_next = &cur->fib6_next;
1898 	}
1899 	return -ENOENT;
1900 }
1901 
1902 /*
1903  *	Tree traversal function.
1904  *
1905  *	Certainly, it is not interrupt safe.
1906  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1907  *	It means, that we can modify tree during walking
1908  *	and use this function for garbage collection, clone pruning,
1909  *	cleaning tree when a device goes down etc. etc.
1910  *
1911  *	It guarantees that every node will be traversed,
1912  *	and that it will be traversed only once.
1913  *
1914  *	Callback function w->func may return:
1915  *	0 -> continue walking.
1916  *	positive value -> walking is suspended (used by tree dumps,
1917  *	and probably by gc, if it will be split to several slices)
1918  *	negative value -> terminate walking.
1919  *
1920  *	The function itself returns:
1921  *	0   -> walk is complete.
1922  *	>0  -> walk is incomplete (i.e. suspended)
1923  *	<0  -> walk is terminated by an error.
1924  *
1925  *	This function is called with tb6_lock held.
1926  */
1927 
1928 static int fib6_walk_continue(struct fib6_walker *w)
1929 {
1930 	struct fib6_node *fn, *pn, *left, *right;
1931 
1932 	/* w->root should always be table->tb6_root */
1933 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1934 
1935 	for (;;) {
1936 		fn = w->node;
1937 		if (!fn)
1938 			return 0;
1939 
1940 		switch (w->state) {
1941 #ifdef CONFIG_IPV6_SUBTREES
1942 		case FWS_S:
1943 			if (FIB6_SUBTREE(fn)) {
1944 				w->node = FIB6_SUBTREE(fn);
1945 				continue;
1946 			}
1947 			w->state = FWS_L;
1948 #endif
1949 			/* fall through */
1950 		case FWS_L:
1951 			left = rcu_dereference_protected(fn->left, 1);
1952 			if (left) {
1953 				w->node = left;
1954 				w->state = FWS_INIT;
1955 				continue;
1956 			}
1957 			w->state = FWS_R;
1958 			/* fall through */
1959 		case FWS_R:
1960 			right = rcu_dereference_protected(fn->right, 1);
1961 			if (right) {
1962 				w->node = right;
1963 				w->state = FWS_INIT;
1964 				continue;
1965 			}
1966 			w->state = FWS_C;
1967 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1968 			/* fall through */
1969 		case FWS_C:
1970 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1971 				int err;
1972 
1973 				if (w->skip) {
1974 					w->skip--;
1975 					goto skip;
1976 				}
1977 
1978 				err = w->func(w);
1979 				if (err)
1980 					return err;
1981 
1982 				w->count++;
1983 				continue;
1984 			}
1985 skip:
1986 			w->state = FWS_U;
1987 			/* fall through */
1988 		case FWS_U:
1989 			if (fn == w->root)
1990 				return 0;
1991 			pn = rcu_dereference_protected(fn->parent, 1);
1992 			left = rcu_dereference_protected(pn->left, 1);
1993 			right = rcu_dereference_protected(pn->right, 1);
1994 			w->node = pn;
1995 #ifdef CONFIG_IPV6_SUBTREES
1996 			if (FIB6_SUBTREE(pn) == fn) {
1997 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
1998 				w->state = FWS_L;
1999 				continue;
2000 			}
2001 #endif
2002 			if (left == fn) {
2003 				w->state = FWS_R;
2004 				continue;
2005 			}
2006 			if (right == fn) {
2007 				w->state = FWS_C;
2008 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2009 				continue;
2010 			}
2011 #if RT6_DEBUG >= 2
2012 			WARN_ON(1);
2013 #endif
2014 		}
2015 	}
2016 }
2017 
2018 static int fib6_walk(struct net *net, struct fib6_walker *w)
2019 {
2020 	int res;
2021 
2022 	w->state = FWS_INIT;
2023 	w->node = w->root;
2024 
2025 	fib6_walker_link(net, w);
2026 	res = fib6_walk_continue(w);
2027 	if (res <= 0)
2028 		fib6_walker_unlink(net, w);
2029 	return res;
2030 }
2031 
2032 static int fib6_clean_node(struct fib6_walker *w)
2033 {
2034 	int res;
2035 	struct fib6_info *rt;
2036 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2037 	struct nl_info info = {
2038 		.nl_net = c->net,
2039 		.skip_notify = c->skip_notify,
2040 	};
2041 
2042 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2043 	    w->node->fn_sernum != c->sernum)
2044 		w->node->fn_sernum = c->sernum;
2045 
2046 	if (!c->func) {
2047 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2048 		w->leaf = NULL;
2049 		return 0;
2050 	}
2051 
2052 	for_each_fib6_walker_rt(w) {
2053 		res = c->func(rt, c->arg);
2054 		if (res == -1) {
2055 			w->leaf = rt;
2056 			res = fib6_del(rt, &info);
2057 			if (res) {
2058 #if RT6_DEBUG >= 2
2059 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2060 					 __func__, rt,
2061 					 rcu_access_pointer(rt->fib6_node),
2062 					 res);
2063 #endif
2064 				continue;
2065 			}
2066 			return 0;
2067 		} else if (res == -2) {
2068 			if (WARN_ON(!rt->fib6_nsiblings))
2069 				continue;
2070 			rt = list_last_entry(&rt->fib6_siblings,
2071 					     struct fib6_info, fib6_siblings);
2072 			continue;
2073 		}
2074 		WARN_ON(res != 0);
2075 	}
2076 	w->leaf = rt;
2077 	return 0;
2078 }
2079 
2080 /*
2081  *	Convenient frontend to tree walker.
2082  *
2083  *	func is called on each route.
2084  *		It may return -2 -> skip multipath route.
2085  *			      -1 -> delete this route.
2086  *		              0  -> continue walking
2087  */
2088 
2089 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2090 			    int (*func)(struct fib6_info *, void *arg),
2091 			    int sernum, void *arg, bool skip_notify)
2092 {
2093 	struct fib6_cleaner c;
2094 
2095 	c.w.root = root;
2096 	c.w.func = fib6_clean_node;
2097 	c.w.count = 0;
2098 	c.w.skip = 0;
2099 	c.func = func;
2100 	c.sernum = sernum;
2101 	c.arg = arg;
2102 	c.net = net;
2103 	c.skip_notify = skip_notify;
2104 
2105 	fib6_walk(net, &c.w);
2106 }
2107 
2108 static void __fib6_clean_all(struct net *net,
2109 			     int (*func)(struct fib6_info *, void *),
2110 			     int sernum, void *arg, bool skip_notify)
2111 {
2112 	struct fib6_table *table;
2113 	struct hlist_head *head;
2114 	unsigned int h;
2115 
2116 	rcu_read_lock();
2117 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2118 		head = &net->ipv6.fib_table_hash[h];
2119 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2120 			spin_lock_bh(&table->tb6_lock);
2121 			fib6_clean_tree(net, &table->tb6_root,
2122 					func, sernum, arg, skip_notify);
2123 			spin_unlock_bh(&table->tb6_lock);
2124 		}
2125 	}
2126 	rcu_read_unlock();
2127 }
2128 
2129 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2130 		    void *arg)
2131 {
2132 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2133 }
2134 
2135 void fib6_clean_all_skip_notify(struct net *net,
2136 				int (*func)(struct fib6_info *, void *),
2137 				void *arg)
2138 {
2139 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2140 }
2141 
2142 static void fib6_flush_trees(struct net *net)
2143 {
2144 	int new_sernum = fib6_new_sernum(net);
2145 
2146 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2147 }
2148 
2149 /*
2150  *	Garbage collection
2151  */
2152 
2153 static int fib6_age(struct fib6_info *rt, void *arg)
2154 {
2155 	struct fib6_gc_args *gc_args = arg;
2156 	unsigned long now = jiffies;
2157 
2158 	/*
2159 	 *	check addrconf expiration here.
2160 	 *	Routes are expired even if they are in use.
2161 	 */
2162 
2163 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2164 		if (time_after(now, rt->expires)) {
2165 			RT6_TRACE("expiring %p\n", rt);
2166 			return -1;
2167 		}
2168 		gc_args->more++;
2169 	}
2170 
2171 	/*	Also age clones in the exception table.
2172 	 *	Note, that clones are aged out
2173 	 *	only if they are not in use now.
2174 	 */
2175 	rt6_age_exceptions(rt, gc_args, now);
2176 
2177 	return 0;
2178 }
2179 
2180 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2181 {
2182 	struct fib6_gc_args gc_args;
2183 	unsigned long now;
2184 
2185 	if (force) {
2186 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2187 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2188 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2189 		return;
2190 	}
2191 	gc_args.timeout = expires ? (int)expires :
2192 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2193 	gc_args.more = 0;
2194 
2195 	fib6_clean_all(net, fib6_age, &gc_args);
2196 	now = jiffies;
2197 	net->ipv6.ip6_rt_last_gc = now;
2198 
2199 	if (gc_args.more)
2200 		mod_timer(&net->ipv6.ip6_fib_timer,
2201 			  round_jiffies(now
2202 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2203 	else
2204 		del_timer(&net->ipv6.ip6_fib_timer);
2205 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2206 }
2207 
2208 static void fib6_gc_timer_cb(struct timer_list *t)
2209 {
2210 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2211 
2212 	fib6_run_gc(0, arg, true);
2213 }
2214 
2215 static int __net_init fib6_net_init(struct net *net)
2216 {
2217 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2218 	int err;
2219 
2220 	err = fib6_notifier_init(net);
2221 	if (err)
2222 		return err;
2223 
2224 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2225 	rwlock_init(&net->ipv6.fib6_walker_lock);
2226 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2227 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2228 
2229 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2230 	if (!net->ipv6.rt6_stats)
2231 		goto out_timer;
2232 
2233 	/* Avoid false sharing : Use at least a full cache line */
2234 	size = max_t(size_t, size, L1_CACHE_BYTES);
2235 
2236 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2237 	if (!net->ipv6.fib_table_hash)
2238 		goto out_rt6_stats;
2239 
2240 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2241 					  GFP_KERNEL);
2242 	if (!net->ipv6.fib6_main_tbl)
2243 		goto out_fib_table_hash;
2244 
2245 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2246 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2247 			   net->ipv6.fib6_null_entry);
2248 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2249 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2250 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2251 
2252 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2253 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2254 					   GFP_KERNEL);
2255 	if (!net->ipv6.fib6_local_tbl)
2256 		goto out_fib6_main_tbl;
2257 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2258 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2259 			   net->ipv6.fib6_null_entry);
2260 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2261 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2262 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2263 #endif
2264 	fib6_tables_init(net);
2265 
2266 	return 0;
2267 
2268 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2269 out_fib6_main_tbl:
2270 	kfree(net->ipv6.fib6_main_tbl);
2271 #endif
2272 out_fib_table_hash:
2273 	kfree(net->ipv6.fib_table_hash);
2274 out_rt6_stats:
2275 	kfree(net->ipv6.rt6_stats);
2276 out_timer:
2277 	fib6_notifier_exit(net);
2278 	return -ENOMEM;
2279 }
2280 
2281 static void fib6_net_exit(struct net *net)
2282 {
2283 	unsigned int i;
2284 
2285 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2286 
2287 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2288 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2289 		struct hlist_node *tmp;
2290 		struct fib6_table *tb;
2291 
2292 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2293 			hlist_del(&tb->tb6_hlist);
2294 			fib6_free_table(tb);
2295 		}
2296 	}
2297 
2298 	kfree(net->ipv6.fib_table_hash);
2299 	kfree(net->ipv6.rt6_stats);
2300 	fib6_notifier_exit(net);
2301 }
2302 
2303 static struct pernet_operations fib6_net_ops = {
2304 	.init = fib6_net_init,
2305 	.exit = fib6_net_exit,
2306 };
2307 
2308 int __init fib6_init(void)
2309 {
2310 	int ret = -ENOMEM;
2311 
2312 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2313 					   sizeof(struct fib6_node),
2314 					   0, SLAB_HWCACHE_ALIGN,
2315 					   NULL);
2316 	if (!fib6_node_kmem)
2317 		goto out;
2318 
2319 	ret = register_pernet_subsys(&fib6_net_ops);
2320 	if (ret)
2321 		goto out_kmem_cache_create;
2322 
2323 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2324 				   inet6_dump_fib, 0);
2325 	if (ret)
2326 		goto out_unregister_subsys;
2327 
2328 	__fib6_flush_trees = fib6_flush_trees;
2329 out:
2330 	return ret;
2331 
2332 out_unregister_subsys:
2333 	unregister_pernet_subsys(&fib6_net_ops);
2334 out_kmem_cache_create:
2335 	kmem_cache_destroy(fib6_node_kmem);
2336 	goto out;
2337 }
2338 
2339 void fib6_gc_cleanup(void)
2340 {
2341 	unregister_pernet_subsys(&fib6_net_ops);
2342 	kmem_cache_destroy(fib6_node_kmem);
2343 }
2344 
2345 #ifdef CONFIG_PROC_FS
2346 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2347 {
2348 	struct fib6_info *rt = v;
2349 	struct ipv6_route_iter *iter = seq->private;
2350 	struct fib6_nh *fib6_nh = rt->fib6_nh;
2351 	unsigned int flags = rt->fib6_flags;
2352 	const struct net_device *dev;
2353 
2354 	if (rt->nh)
2355 		fib6_nh = nexthop_fib6_nh(rt->nh);
2356 
2357 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2358 
2359 #ifdef CONFIG_IPV6_SUBTREES
2360 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2361 #else
2362 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2363 #endif
2364 	if (fib6_nh->fib_nh_gw_family) {
2365 		flags |= RTF_GATEWAY;
2366 		seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2367 	} else {
2368 		seq_puts(seq, "00000000000000000000000000000000");
2369 	}
2370 
2371 	dev = fib6_nh->fib_nh_dev;
2372 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2373 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2374 		   flags, dev ? dev->name : "");
2375 	iter->w.leaf = NULL;
2376 	return 0;
2377 }
2378 
2379 static int ipv6_route_yield(struct fib6_walker *w)
2380 {
2381 	struct ipv6_route_iter *iter = w->args;
2382 
2383 	if (!iter->skip)
2384 		return 1;
2385 
2386 	do {
2387 		iter->w.leaf = rcu_dereference_protected(
2388 				iter->w.leaf->fib6_next,
2389 				lockdep_is_held(&iter->tbl->tb6_lock));
2390 		iter->skip--;
2391 		if (!iter->skip && iter->w.leaf)
2392 			return 1;
2393 	} while (iter->w.leaf);
2394 
2395 	return 0;
2396 }
2397 
2398 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2399 				      struct net *net)
2400 {
2401 	memset(&iter->w, 0, sizeof(iter->w));
2402 	iter->w.func = ipv6_route_yield;
2403 	iter->w.root = &iter->tbl->tb6_root;
2404 	iter->w.state = FWS_INIT;
2405 	iter->w.node = iter->w.root;
2406 	iter->w.args = iter;
2407 	iter->sernum = iter->w.root->fn_sernum;
2408 	INIT_LIST_HEAD(&iter->w.lh);
2409 	fib6_walker_link(net, &iter->w);
2410 }
2411 
2412 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2413 						    struct net *net)
2414 {
2415 	unsigned int h;
2416 	struct hlist_node *node;
2417 
2418 	if (tbl) {
2419 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2420 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2421 	} else {
2422 		h = 0;
2423 		node = NULL;
2424 	}
2425 
2426 	while (!node && h < FIB6_TABLE_HASHSZ) {
2427 		node = rcu_dereference_bh(
2428 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2429 	}
2430 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2431 }
2432 
2433 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2434 {
2435 	if (iter->sernum != iter->w.root->fn_sernum) {
2436 		iter->sernum = iter->w.root->fn_sernum;
2437 		iter->w.state = FWS_INIT;
2438 		iter->w.node = iter->w.root;
2439 		WARN_ON(iter->w.skip);
2440 		iter->w.skip = iter->w.count;
2441 	}
2442 }
2443 
2444 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2445 {
2446 	int r;
2447 	struct fib6_info *n;
2448 	struct net *net = seq_file_net(seq);
2449 	struct ipv6_route_iter *iter = seq->private;
2450 
2451 	if (!v)
2452 		goto iter_table;
2453 
2454 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2455 	if (n) {
2456 		++*pos;
2457 		return n;
2458 	}
2459 
2460 iter_table:
2461 	ipv6_route_check_sernum(iter);
2462 	spin_lock_bh(&iter->tbl->tb6_lock);
2463 	r = fib6_walk_continue(&iter->w);
2464 	spin_unlock_bh(&iter->tbl->tb6_lock);
2465 	if (r > 0) {
2466 		if (v)
2467 			++*pos;
2468 		return iter->w.leaf;
2469 	} else if (r < 0) {
2470 		fib6_walker_unlink(net, &iter->w);
2471 		return NULL;
2472 	}
2473 	fib6_walker_unlink(net, &iter->w);
2474 
2475 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2476 	if (!iter->tbl)
2477 		return NULL;
2478 
2479 	ipv6_route_seq_setup_walk(iter, net);
2480 	goto iter_table;
2481 }
2482 
2483 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2484 	__acquires(RCU_BH)
2485 {
2486 	struct net *net = seq_file_net(seq);
2487 	struct ipv6_route_iter *iter = seq->private;
2488 
2489 	rcu_read_lock_bh();
2490 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2491 	iter->skip = *pos;
2492 
2493 	if (iter->tbl) {
2494 		ipv6_route_seq_setup_walk(iter, net);
2495 		return ipv6_route_seq_next(seq, NULL, pos);
2496 	} else {
2497 		return NULL;
2498 	}
2499 }
2500 
2501 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2502 {
2503 	struct fib6_walker *w = &iter->w;
2504 	return w->node && !(w->state == FWS_U && w->node == w->root);
2505 }
2506 
2507 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2508 	__releases(RCU_BH)
2509 {
2510 	struct net *net = seq_file_net(seq);
2511 	struct ipv6_route_iter *iter = seq->private;
2512 
2513 	if (ipv6_route_iter_active(iter))
2514 		fib6_walker_unlink(net, &iter->w);
2515 
2516 	rcu_read_unlock_bh();
2517 }
2518 
2519 const struct seq_operations ipv6_route_seq_ops = {
2520 	.start	= ipv6_route_seq_start,
2521 	.next	= ipv6_route_seq_next,
2522 	.stop	= ipv6_route_seq_stop,
2523 	.show	= ipv6_route_seq_show
2524 };
2525 #endif /* CONFIG_PROC_FS */
2526