xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 005f3e67)
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,
361 				    enum fib_event_type event_type,
362 				    struct fib6_info *rt,
363 				    struct netlink_ext_ack *extack)
364 {
365 	struct fib6_entry_notifier_info info = {
366 		.info.extack = extack,
367 		.rt = rt,
368 	};
369 
370 	return call_fib6_notifier(nb, event_type, &info.info);
371 }
372 
373 int call_fib6_entry_notifiers(struct net *net,
374 			      enum fib_event_type event_type,
375 			      struct fib6_info *rt,
376 			      struct netlink_ext_ack *extack)
377 {
378 	struct fib6_entry_notifier_info info = {
379 		.info.extack = extack,
380 		.rt = rt,
381 	};
382 
383 	rt->fib6_table->fib_seq++;
384 	return call_fib6_notifiers(net, event_type, &info.info);
385 }
386 
387 int call_fib6_multipath_entry_notifiers(struct net *net,
388 					enum fib_event_type event_type,
389 					struct fib6_info *rt,
390 					unsigned int nsiblings,
391 					struct netlink_ext_ack *extack)
392 {
393 	struct fib6_entry_notifier_info info = {
394 		.info.extack = extack,
395 		.rt = rt,
396 		.nsiblings = nsiblings,
397 	};
398 
399 	rt->fib6_table->fib_seq++;
400 	return call_fib6_notifiers(net, event_type, &info.info);
401 }
402 
403 struct fib6_dump_arg {
404 	struct net *net;
405 	struct notifier_block *nb;
406 	struct netlink_ext_ack *extack;
407 };
408 
409 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
410 {
411 	if (rt == arg->net->ipv6.fib6_null_entry)
412 		return 0;
413 	return call_fib6_entry_notifier(arg->nb, FIB_EVENT_ENTRY_ADD,
414 					rt, arg->extack);
415 }
416 
417 static int fib6_node_dump(struct fib6_walker *w)
418 {
419 	struct fib6_info *rt;
420 	int err = 0;
421 
422 	for_each_fib6_walker_rt(w) {
423 		err = fib6_rt_dump(rt, w->args);
424 		if (err)
425 			break;
426 	}
427 	w->leaf = NULL;
428 	return err;
429 }
430 
431 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
432 			   struct fib6_walker *w)
433 {
434 	int err;
435 
436 	w->root = &tb->tb6_root;
437 	spin_lock_bh(&tb->tb6_lock);
438 	err = fib6_walk(net, w);
439 	spin_unlock_bh(&tb->tb6_lock);
440 	return err;
441 }
442 
443 /* Called with rcu_read_lock() */
444 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
445 		     struct netlink_ext_ack *extack)
446 {
447 	struct fib6_dump_arg arg;
448 	struct fib6_walker *w;
449 	unsigned int h;
450 	int err = 0;
451 
452 	w = kzalloc(sizeof(*w), GFP_ATOMIC);
453 	if (!w)
454 		return -ENOMEM;
455 
456 	w->func = fib6_node_dump;
457 	arg.net = net;
458 	arg.nb = nb;
459 	arg.extack = extack;
460 	w->args = &arg;
461 
462 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
463 		struct hlist_head *head = &net->ipv6.fib_table_hash[h];
464 		struct fib6_table *tb;
465 
466 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
467 			err = fib6_table_dump(net, tb, w);
468 			if (err < 0)
469 				goto out;
470 		}
471 	}
472 
473 out:
474 	kfree(w);
475 
476 	return err;
477 }
478 
479 static int fib6_dump_node(struct fib6_walker *w)
480 {
481 	int res;
482 	struct fib6_info *rt;
483 
484 	for_each_fib6_walker_rt(w) {
485 		res = rt6_dump_route(rt, w->args, w->skip_in_node);
486 		if (res >= 0) {
487 			/* Frame is full, suspend walking */
488 			w->leaf = rt;
489 
490 			/* We'll restart from this node, so if some routes were
491 			 * already dumped, skip them next time.
492 			 */
493 			w->skip_in_node += res;
494 
495 			return 1;
496 		}
497 		w->skip_in_node = 0;
498 
499 		/* Multipath routes are dumped in one route with the
500 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
501 		 * last sibling of this route (no need to dump the
502 		 * sibling routes again)
503 		 */
504 		if (rt->fib6_nsiblings)
505 			rt = list_last_entry(&rt->fib6_siblings,
506 					     struct fib6_info,
507 					     fib6_siblings);
508 	}
509 	w->leaf = NULL;
510 	return 0;
511 }
512 
513 static void fib6_dump_end(struct netlink_callback *cb)
514 {
515 	struct net *net = sock_net(cb->skb->sk);
516 	struct fib6_walker *w = (void *)cb->args[2];
517 
518 	if (w) {
519 		if (cb->args[4]) {
520 			cb->args[4] = 0;
521 			fib6_walker_unlink(net, w);
522 		}
523 		cb->args[2] = 0;
524 		kfree(w);
525 	}
526 	cb->done = (void *)cb->args[3];
527 	cb->args[1] = 3;
528 }
529 
530 static int fib6_dump_done(struct netlink_callback *cb)
531 {
532 	fib6_dump_end(cb);
533 	return cb->done ? cb->done(cb) : 0;
534 }
535 
536 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
537 			   struct netlink_callback *cb)
538 {
539 	struct net *net = sock_net(skb->sk);
540 	struct fib6_walker *w;
541 	int res;
542 
543 	w = (void *)cb->args[2];
544 	w->root = &table->tb6_root;
545 
546 	if (cb->args[4] == 0) {
547 		w->count = 0;
548 		w->skip = 0;
549 		w->skip_in_node = 0;
550 
551 		spin_lock_bh(&table->tb6_lock);
552 		res = fib6_walk(net, w);
553 		spin_unlock_bh(&table->tb6_lock);
554 		if (res > 0) {
555 			cb->args[4] = 1;
556 			cb->args[5] = w->root->fn_sernum;
557 		}
558 	} else {
559 		if (cb->args[5] != w->root->fn_sernum) {
560 			/* Begin at the root if the tree changed */
561 			cb->args[5] = w->root->fn_sernum;
562 			w->state = FWS_INIT;
563 			w->node = w->root;
564 			w->skip = w->count;
565 			w->skip_in_node = 0;
566 		} else
567 			w->skip = 0;
568 
569 		spin_lock_bh(&table->tb6_lock);
570 		res = fib6_walk_continue(w);
571 		spin_unlock_bh(&table->tb6_lock);
572 		if (res <= 0) {
573 			fib6_walker_unlink(net, w);
574 			cb->args[4] = 0;
575 		}
576 	}
577 
578 	return res;
579 }
580 
581 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
582 {
583 	struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
584 					 .filter.dump_routes = true };
585 	const struct nlmsghdr *nlh = cb->nlh;
586 	struct net *net = sock_net(skb->sk);
587 	unsigned int h, s_h;
588 	unsigned int e = 0, s_e;
589 	struct fib6_walker *w;
590 	struct fib6_table *tb;
591 	struct hlist_head *head;
592 	int res = 0;
593 
594 	if (cb->strict_check) {
595 		int err;
596 
597 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
598 		if (err < 0)
599 			return err;
600 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
601 		struct rtmsg *rtm = nlmsg_data(nlh);
602 
603 		if (rtm->rtm_flags & RTM_F_PREFIX)
604 			arg.filter.flags = RTM_F_PREFIX;
605 	}
606 
607 	w = (void *)cb->args[2];
608 	if (!w) {
609 		/* New dump:
610 		 *
611 		 * 1. hook callback destructor.
612 		 */
613 		cb->args[3] = (long)cb->done;
614 		cb->done = fib6_dump_done;
615 
616 		/*
617 		 * 2. allocate and initialize walker.
618 		 */
619 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
620 		if (!w)
621 			return -ENOMEM;
622 		w->func = fib6_dump_node;
623 		cb->args[2] = (long)w;
624 	}
625 
626 	arg.skb = skb;
627 	arg.cb = cb;
628 	arg.net = net;
629 	w->args = &arg;
630 
631 	if (arg.filter.table_id) {
632 		tb = fib6_get_table(net, arg.filter.table_id);
633 		if (!tb) {
634 			if (arg.filter.dump_all_families)
635 				goto out;
636 
637 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
638 			return -ENOENT;
639 		}
640 
641 		if (!cb->args[0]) {
642 			res = fib6_dump_table(tb, skb, cb);
643 			if (!res)
644 				cb->args[0] = 1;
645 		}
646 		goto out;
647 	}
648 
649 	s_h = cb->args[0];
650 	s_e = cb->args[1];
651 
652 	rcu_read_lock();
653 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
654 		e = 0;
655 		head = &net->ipv6.fib_table_hash[h];
656 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
657 			if (e < s_e)
658 				goto next;
659 			res = fib6_dump_table(tb, skb, cb);
660 			if (res != 0)
661 				goto out_unlock;
662 next:
663 			e++;
664 		}
665 	}
666 out_unlock:
667 	rcu_read_unlock();
668 	cb->args[1] = e;
669 	cb->args[0] = h;
670 out:
671 	res = res < 0 ? res : skb->len;
672 	if (res <= 0)
673 		fib6_dump_end(cb);
674 	return res;
675 }
676 
677 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
678 {
679 	if (!f6i)
680 		return;
681 
682 	if (f6i->fib6_metrics == &dst_default_metrics) {
683 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
684 
685 		if (!p)
686 			return;
687 
688 		refcount_set(&p->refcnt, 1);
689 		f6i->fib6_metrics = p;
690 	}
691 
692 	f6i->fib6_metrics->metrics[metric - 1] = val;
693 }
694 
695 /*
696  *	Routing Table
697  *
698  *	return the appropriate node for a routing tree "add" operation
699  *	by either creating and inserting or by returning an existing
700  *	node.
701  */
702 
703 static struct fib6_node *fib6_add_1(struct net *net,
704 				    struct fib6_table *table,
705 				    struct fib6_node *root,
706 				    struct in6_addr *addr, int plen,
707 				    int offset, int allow_create,
708 				    int replace_required,
709 				    struct netlink_ext_ack *extack)
710 {
711 	struct fib6_node *fn, *in, *ln;
712 	struct fib6_node *pn = NULL;
713 	struct rt6key *key;
714 	int	bit;
715 	__be32	dir = 0;
716 
717 	RT6_TRACE("fib6_add_1\n");
718 
719 	/* insert node in tree */
720 
721 	fn = root;
722 
723 	do {
724 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
725 					    lockdep_is_held(&table->tb6_lock));
726 		key = (struct rt6key *)((u8 *)leaf + offset);
727 
728 		/*
729 		 *	Prefix match
730 		 */
731 		if (plen < fn->fn_bit ||
732 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
733 			if (!allow_create) {
734 				if (replace_required) {
735 					NL_SET_ERR_MSG(extack,
736 						       "Can not replace route - no match found");
737 					pr_warn("Can't replace route, no match found\n");
738 					return ERR_PTR(-ENOENT);
739 				}
740 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
741 			}
742 			goto insert_above;
743 		}
744 
745 		/*
746 		 *	Exact match ?
747 		 */
748 
749 		if (plen == fn->fn_bit) {
750 			/* clean up an intermediate node */
751 			if (!(fn->fn_flags & RTN_RTINFO)) {
752 				RCU_INIT_POINTER(fn->leaf, NULL);
753 				fib6_info_release(leaf);
754 			/* remove null_entry in the root node */
755 			} else if (fn->fn_flags & RTN_TL_ROOT &&
756 				   rcu_access_pointer(fn->leaf) ==
757 				   net->ipv6.fib6_null_entry) {
758 				RCU_INIT_POINTER(fn->leaf, NULL);
759 			}
760 
761 			return fn;
762 		}
763 
764 		/*
765 		 *	We have more bits to go
766 		 */
767 
768 		/* Try to walk down on tree. */
769 		dir = addr_bit_set(addr, fn->fn_bit);
770 		pn = fn;
771 		fn = dir ?
772 		     rcu_dereference_protected(fn->right,
773 					lockdep_is_held(&table->tb6_lock)) :
774 		     rcu_dereference_protected(fn->left,
775 					lockdep_is_held(&table->tb6_lock));
776 	} while (fn);
777 
778 	if (!allow_create) {
779 		/* We should not create new node because
780 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
781 		 * I assume it is safe to require NLM_F_CREATE when
782 		 * REPLACE flag is used! Later we may want to remove the
783 		 * check for replace_required, because according
784 		 * to netlink specification, NLM_F_CREATE
785 		 * MUST be specified if new route is created.
786 		 * That would keep IPv6 consistent with IPv4
787 		 */
788 		if (replace_required) {
789 			NL_SET_ERR_MSG(extack,
790 				       "Can not replace route - no match found");
791 			pr_warn("Can't replace route, no match found\n");
792 			return ERR_PTR(-ENOENT);
793 		}
794 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
795 	}
796 	/*
797 	 *	We walked to the bottom of tree.
798 	 *	Create new leaf node without children.
799 	 */
800 
801 	ln = node_alloc(net);
802 
803 	if (!ln)
804 		return ERR_PTR(-ENOMEM);
805 	ln->fn_bit = plen;
806 	RCU_INIT_POINTER(ln->parent, pn);
807 
808 	if (dir)
809 		rcu_assign_pointer(pn->right, ln);
810 	else
811 		rcu_assign_pointer(pn->left, ln);
812 
813 	return ln;
814 
815 
816 insert_above:
817 	/*
818 	 * split since we don't have a common prefix anymore or
819 	 * we have a less significant route.
820 	 * we've to insert an intermediate node on the list
821 	 * this new node will point to the one we need to create
822 	 * and the current
823 	 */
824 
825 	pn = rcu_dereference_protected(fn->parent,
826 				       lockdep_is_held(&table->tb6_lock));
827 
828 	/* find 1st bit in difference between the 2 addrs.
829 
830 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
831 	   but if it is >= plen, the value is ignored in any case.
832 	 */
833 
834 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
835 
836 	/*
837 	 *		(intermediate)[in]
838 	 *	          /	   \
839 	 *	(new leaf node)[ln] (old node)[fn]
840 	 */
841 	if (plen > bit) {
842 		in = node_alloc(net);
843 		ln = node_alloc(net);
844 
845 		if (!in || !ln) {
846 			if (in)
847 				node_free_immediate(net, in);
848 			if (ln)
849 				node_free_immediate(net, ln);
850 			return ERR_PTR(-ENOMEM);
851 		}
852 
853 		/*
854 		 * new intermediate node.
855 		 * RTN_RTINFO will
856 		 * be off since that an address that chooses one of
857 		 * the branches would not match less specific routes
858 		 * in the other branch
859 		 */
860 
861 		in->fn_bit = bit;
862 
863 		RCU_INIT_POINTER(in->parent, pn);
864 		in->leaf = fn->leaf;
865 		fib6_info_hold(rcu_dereference_protected(in->leaf,
866 				lockdep_is_held(&table->tb6_lock)));
867 
868 		/* update parent pointer */
869 		if (dir)
870 			rcu_assign_pointer(pn->right, in);
871 		else
872 			rcu_assign_pointer(pn->left, in);
873 
874 		ln->fn_bit = plen;
875 
876 		RCU_INIT_POINTER(ln->parent, in);
877 		rcu_assign_pointer(fn->parent, in);
878 
879 		if (addr_bit_set(addr, bit)) {
880 			rcu_assign_pointer(in->right, ln);
881 			rcu_assign_pointer(in->left, fn);
882 		} else {
883 			rcu_assign_pointer(in->left, ln);
884 			rcu_assign_pointer(in->right, fn);
885 		}
886 	} else { /* plen <= bit */
887 
888 		/*
889 		 *		(new leaf node)[ln]
890 		 *	          /	   \
891 		 *	     (old node)[fn] NULL
892 		 */
893 
894 		ln = node_alloc(net);
895 
896 		if (!ln)
897 			return ERR_PTR(-ENOMEM);
898 
899 		ln->fn_bit = plen;
900 
901 		RCU_INIT_POINTER(ln->parent, pn);
902 
903 		if (addr_bit_set(&key->addr, plen))
904 			RCU_INIT_POINTER(ln->right, fn);
905 		else
906 			RCU_INIT_POINTER(ln->left, fn);
907 
908 		rcu_assign_pointer(fn->parent, ln);
909 
910 		if (dir)
911 			rcu_assign_pointer(pn->right, ln);
912 		else
913 			rcu_assign_pointer(pn->left, ln);
914 	}
915 	return ln;
916 }
917 
918 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
919 				  const struct fib6_info *match,
920 				  const struct fib6_table *table)
921 {
922 	int cpu;
923 
924 	if (!fib6_nh->rt6i_pcpu)
925 		return;
926 
927 	/* release the reference to this fib entry from
928 	 * all of its cached pcpu routes
929 	 */
930 	for_each_possible_cpu(cpu) {
931 		struct rt6_info **ppcpu_rt;
932 		struct rt6_info *pcpu_rt;
933 
934 		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
935 		pcpu_rt = *ppcpu_rt;
936 
937 		/* only dropping the 'from' reference if the cached route
938 		 * is using 'match'. The cached pcpu_rt->from only changes
939 		 * from a fib6_info to NULL (ip6_dst_destroy); it can never
940 		 * change from one fib6_info reference to another
941 		 */
942 		if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
943 			struct fib6_info *from;
944 
945 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
946 			fib6_info_release(from);
947 		}
948 	}
949 }
950 
951 struct fib6_nh_pcpu_arg {
952 	struct fib6_info	*from;
953 	const struct fib6_table *table;
954 };
955 
956 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
957 {
958 	struct fib6_nh_pcpu_arg *arg = _arg;
959 
960 	__fib6_drop_pcpu_from(nh, arg->from, arg->table);
961 	return 0;
962 }
963 
964 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
965 				const struct fib6_table *table)
966 {
967 	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
968 	 * while we are cleaning them here.
969 	 */
970 	f6i->fib6_destroying = 1;
971 	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
972 
973 	if (f6i->nh) {
974 		struct fib6_nh_pcpu_arg arg = {
975 			.from = f6i,
976 			.table = table
977 		};
978 
979 		nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
980 					 &arg);
981 	} else {
982 		struct fib6_nh *fib6_nh;
983 
984 		fib6_nh = f6i->fib6_nh;
985 		__fib6_drop_pcpu_from(fib6_nh, f6i, table);
986 	}
987 }
988 
989 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
990 			  struct net *net)
991 {
992 	struct fib6_table *table = rt->fib6_table;
993 
994 	fib6_drop_pcpu_from(rt, table);
995 
996 	if (rt->nh && !list_empty(&rt->nh_list))
997 		list_del_init(&rt->nh_list);
998 
999 	if (refcount_read(&rt->fib6_ref) != 1) {
1000 		/* This route is used as dummy address holder in some split
1001 		 * nodes. It is not leaked, but it still holds other resources,
1002 		 * which must be released in time. So, scan ascendant nodes
1003 		 * and replace dummy references to this route with references
1004 		 * to still alive ones.
1005 		 */
1006 		while (fn) {
1007 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1008 					    lockdep_is_held(&table->tb6_lock));
1009 			struct fib6_info *new_leaf;
1010 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1011 				new_leaf = fib6_find_prefix(net, table, fn);
1012 				fib6_info_hold(new_leaf);
1013 
1014 				rcu_assign_pointer(fn->leaf, new_leaf);
1015 				fib6_info_release(rt);
1016 			}
1017 			fn = rcu_dereference_protected(fn->parent,
1018 				    lockdep_is_held(&table->tb6_lock));
1019 		}
1020 	}
1021 }
1022 
1023 /*
1024  *	Insert routing information in a node.
1025  */
1026 
1027 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1028 			    struct nl_info *info,
1029 			    struct netlink_ext_ack *extack)
1030 {
1031 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1032 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1033 	struct fib6_info *iter = NULL;
1034 	struct fib6_info __rcu **ins;
1035 	struct fib6_info __rcu **fallback_ins = NULL;
1036 	int replace = (info->nlh &&
1037 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1038 	int add = (!info->nlh ||
1039 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
1040 	int found = 0;
1041 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1042 	u16 nlflags = NLM_F_EXCL;
1043 	int err;
1044 
1045 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1046 		nlflags |= NLM_F_APPEND;
1047 
1048 	ins = &fn->leaf;
1049 
1050 	for (iter = leaf; iter;
1051 	     iter = rcu_dereference_protected(iter->fib6_next,
1052 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1053 		/*
1054 		 *	Search for duplicates
1055 		 */
1056 
1057 		if (iter->fib6_metric == rt->fib6_metric) {
1058 			/*
1059 			 *	Same priority level
1060 			 */
1061 			if (info->nlh &&
1062 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1063 				return -EEXIST;
1064 
1065 			nlflags &= ~NLM_F_EXCL;
1066 			if (replace) {
1067 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1068 					found++;
1069 					break;
1070 				}
1071 				if (rt_can_ecmp)
1072 					fallback_ins = fallback_ins ?: ins;
1073 				goto next_iter;
1074 			}
1075 
1076 			if (rt6_duplicate_nexthop(iter, rt)) {
1077 				if (rt->fib6_nsiblings)
1078 					rt->fib6_nsiblings = 0;
1079 				if (!(iter->fib6_flags & RTF_EXPIRES))
1080 					return -EEXIST;
1081 				if (!(rt->fib6_flags & RTF_EXPIRES))
1082 					fib6_clean_expires(iter);
1083 				else
1084 					fib6_set_expires(iter, rt->expires);
1085 
1086 				if (rt->fib6_pmtu)
1087 					fib6_metric_set(iter, RTAX_MTU,
1088 							rt->fib6_pmtu);
1089 				return -EEXIST;
1090 			}
1091 			/* If we have the same destination and the same metric,
1092 			 * but not the same gateway, then the route we try to
1093 			 * add is sibling to this route, increment our counter
1094 			 * of siblings, and later we will add our route to the
1095 			 * list.
1096 			 * Only static routes (which don't have flag
1097 			 * RTF_EXPIRES) are used for ECMPv6.
1098 			 *
1099 			 * To avoid long list, we only had siblings if the
1100 			 * route have a gateway.
1101 			 */
1102 			if (rt_can_ecmp &&
1103 			    rt6_qualify_for_ecmp(iter))
1104 				rt->fib6_nsiblings++;
1105 		}
1106 
1107 		if (iter->fib6_metric > rt->fib6_metric)
1108 			break;
1109 
1110 next_iter:
1111 		ins = &iter->fib6_next;
1112 	}
1113 
1114 	if (fallback_ins && !found) {
1115 		/* No ECMP-able route found, replace first non-ECMP one */
1116 		ins = fallback_ins;
1117 		iter = rcu_dereference_protected(*ins,
1118 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1119 		found++;
1120 	}
1121 
1122 	/* Reset round-robin state, if necessary */
1123 	if (ins == &fn->leaf)
1124 		fn->rr_ptr = NULL;
1125 
1126 	/* Link this route to others same route. */
1127 	if (rt->fib6_nsiblings) {
1128 		unsigned int fib6_nsiblings;
1129 		struct fib6_info *sibling, *temp_sibling;
1130 
1131 		/* Find the first route that have the same metric */
1132 		sibling = leaf;
1133 		while (sibling) {
1134 			if (sibling->fib6_metric == rt->fib6_metric &&
1135 			    rt6_qualify_for_ecmp(sibling)) {
1136 				list_add_tail(&rt->fib6_siblings,
1137 					      &sibling->fib6_siblings);
1138 				break;
1139 			}
1140 			sibling = rcu_dereference_protected(sibling->fib6_next,
1141 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1142 		}
1143 		/* For each sibling in the list, increment the counter of
1144 		 * siblings. BUG() if counters does not match, list of siblings
1145 		 * is broken!
1146 		 */
1147 		fib6_nsiblings = 0;
1148 		list_for_each_entry_safe(sibling, temp_sibling,
1149 					 &rt->fib6_siblings, fib6_siblings) {
1150 			sibling->fib6_nsiblings++;
1151 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1152 			fib6_nsiblings++;
1153 		}
1154 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1155 		rt6_multipath_rebalance(temp_sibling);
1156 	}
1157 
1158 	/*
1159 	 *	insert node
1160 	 */
1161 	if (!replace) {
1162 		if (!add)
1163 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1164 
1165 add:
1166 		nlflags |= NLM_F_CREATE;
1167 
1168 		if (!info->skip_notify_kernel) {
1169 			err = call_fib6_entry_notifiers(info->nl_net,
1170 							FIB_EVENT_ENTRY_ADD,
1171 							rt, extack);
1172 			if (err) {
1173 				struct fib6_info *sibling, *next_sibling;
1174 
1175 				/* If the route has siblings, then it first
1176 				 * needs to be unlinked from them.
1177 				 */
1178 				if (!rt->fib6_nsiblings)
1179 					return err;
1180 
1181 				list_for_each_entry_safe(sibling, next_sibling,
1182 							 &rt->fib6_siblings,
1183 							 fib6_siblings)
1184 					sibling->fib6_nsiblings--;
1185 				rt->fib6_nsiblings = 0;
1186 				list_del_init(&rt->fib6_siblings);
1187 				rt6_multipath_rebalance(next_sibling);
1188 				return err;
1189 			}
1190 		}
1191 
1192 		rcu_assign_pointer(rt->fib6_next, iter);
1193 		fib6_info_hold(rt);
1194 		rcu_assign_pointer(rt->fib6_node, fn);
1195 		rcu_assign_pointer(*ins, rt);
1196 		if (!info->skip_notify)
1197 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1198 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1199 
1200 		if (!(fn->fn_flags & RTN_RTINFO)) {
1201 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1202 			fn->fn_flags |= RTN_RTINFO;
1203 		}
1204 
1205 	} else {
1206 		int nsiblings;
1207 
1208 		if (!found) {
1209 			if (add)
1210 				goto add;
1211 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1212 			return -ENOENT;
1213 		}
1214 
1215 		if (!info->skip_notify_kernel) {
1216 			err = call_fib6_entry_notifiers(info->nl_net,
1217 							FIB_EVENT_ENTRY_REPLACE,
1218 							rt, extack);
1219 			if (err)
1220 				return err;
1221 		}
1222 
1223 		fib6_info_hold(rt);
1224 		rcu_assign_pointer(rt->fib6_node, fn);
1225 		rt->fib6_next = iter->fib6_next;
1226 		rcu_assign_pointer(*ins, rt);
1227 		if (!info->skip_notify)
1228 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1229 		if (!(fn->fn_flags & RTN_RTINFO)) {
1230 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1231 			fn->fn_flags |= RTN_RTINFO;
1232 		}
1233 		nsiblings = iter->fib6_nsiblings;
1234 		iter->fib6_node = NULL;
1235 		fib6_purge_rt(iter, fn, info->nl_net);
1236 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1237 			fn->rr_ptr = NULL;
1238 		fib6_info_release(iter);
1239 
1240 		if (nsiblings) {
1241 			/* Replacing an ECMP route, remove all siblings */
1242 			ins = &rt->fib6_next;
1243 			iter = rcu_dereference_protected(*ins,
1244 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1245 			while (iter) {
1246 				if (iter->fib6_metric > rt->fib6_metric)
1247 					break;
1248 				if (rt6_qualify_for_ecmp(iter)) {
1249 					*ins = iter->fib6_next;
1250 					iter->fib6_node = NULL;
1251 					fib6_purge_rt(iter, fn, info->nl_net);
1252 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1253 						fn->rr_ptr = NULL;
1254 					fib6_info_release(iter);
1255 					nsiblings--;
1256 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1257 				} else {
1258 					ins = &iter->fib6_next;
1259 				}
1260 				iter = rcu_dereference_protected(*ins,
1261 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1262 			}
1263 			WARN_ON(nsiblings != 0);
1264 		}
1265 	}
1266 
1267 	return 0;
1268 }
1269 
1270 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1271 {
1272 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1273 	    (rt->fib6_flags & RTF_EXPIRES))
1274 		mod_timer(&net->ipv6.ip6_fib_timer,
1275 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1276 }
1277 
1278 void fib6_force_start_gc(struct net *net)
1279 {
1280 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1281 		mod_timer(&net->ipv6.ip6_fib_timer,
1282 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1283 }
1284 
1285 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1286 					   int sernum)
1287 {
1288 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1289 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1290 
1291 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1292 	smp_wmb();
1293 	while (fn) {
1294 		fn->fn_sernum = sernum;
1295 		fn = rcu_dereference_protected(fn->parent,
1296 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1297 	}
1298 }
1299 
1300 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1301 {
1302 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1303 }
1304 
1305 /* allow ipv4 to update sernum via ipv6_stub */
1306 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1307 {
1308 	spin_lock_bh(&f6i->fib6_table->tb6_lock);
1309 	fib6_update_sernum_upto_root(net, f6i);
1310 	spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1311 }
1312 
1313 /*
1314  *	Add routing information to the routing tree.
1315  *	<destination addr>/<source addr>
1316  *	with source addr info in sub-trees
1317  *	Need to own table->tb6_lock
1318  */
1319 
1320 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1321 	     struct nl_info *info, struct netlink_ext_ack *extack)
1322 {
1323 	struct fib6_table *table = rt->fib6_table;
1324 	struct fib6_node *fn, *pn = NULL;
1325 	int err = -ENOMEM;
1326 	int allow_create = 1;
1327 	int replace_required = 0;
1328 	int sernum = fib6_new_sernum(info->nl_net);
1329 
1330 	if (info->nlh) {
1331 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1332 			allow_create = 0;
1333 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1334 			replace_required = 1;
1335 	}
1336 	if (!allow_create && !replace_required)
1337 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1338 
1339 	fn = fib6_add_1(info->nl_net, table, root,
1340 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1341 			offsetof(struct fib6_info, fib6_dst), allow_create,
1342 			replace_required, extack);
1343 	if (IS_ERR(fn)) {
1344 		err = PTR_ERR(fn);
1345 		fn = NULL;
1346 		goto out;
1347 	}
1348 
1349 	pn = fn;
1350 
1351 #ifdef CONFIG_IPV6_SUBTREES
1352 	if (rt->fib6_src.plen) {
1353 		struct fib6_node *sn;
1354 
1355 		if (!rcu_access_pointer(fn->subtree)) {
1356 			struct fib6_node *sfn;
1357 
1358 			/*
1359 			 * Create subtree.
1360 			 *
1361 			 *		fn[main tree]
1362 			 *		|
1363 			 *		sfn[subtree root]
1364 			 *		   \
1365 			 *		    sn[new leaf node]
1366 			 */
1367 
1368 			/* Create subtree root node */
1369 			sfn = node_alloc(info->nl_net);
1370 			if (!sfn)
1371 				goto failure;
1372 
1373 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1374 			rcu_assign_pointer(sfn->leaf,
1375 					   info->nl_net->ipv6.fib6_null_entry);
1376 			sfn->fn_flags = RTN_ROOT;
1377 
1378 			/* Now add the first leaf node to new subtree */
1379 
1380 			sn = fib6_add_1(info->nl_net, table, sfn,
1381 					&rt->fib6_src.addr, rt->fib6_src.plen,
1382 					offsetof(struct fib6_info, fib6_src),
1383 					allow_create, replace_required, extack);
1384 
1385 			if (IS_ERR(sn)) {
1386 				/* If it is failed, discard just allocated
1387 				   root, and then (in failure) stale node
1388 				   in main tree.
1389 				 */
1390 				node_free_immediate(info->nl_net, sfn);
1391 				err = PTR_ERR(sn);
1392 				goto failure;
1393 			}
1394 
1395 			/* Now link new subtree to main tree */
1396 			rcu_assign_pointer(sfn->parent, fn);
1397 			rcu_assign_pointer(fn->subtree, sfn);
1398 		} else {
1399 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1400 					&rt->fib6_src.addr, rt->fib6_src.plen,
1401 					offsetof(struct fib6_info, fib6_src),
1402 					allow_create, replace_required, extack);
1403 
1404 			if (IS_ERR(sn)) {
1405 				err = PTR_ERR(sn);
1406 				goto failure;
1407 			}
1408 		}
1409 
1410 		if (!rcu_access_pointer(fn->leaf)) {
1411 			if (fn->fn_flags & RTN_TL_ROOT) {
1412 				/* put back null_entry for root node */
1413 				rcu_assign_pointer(fn->leaf,
1414 					    info->nl_net->ipv6.fib6_null_entry);
1415 			} else {
1416 				fib6_info_hold(rt);
1417 				rcu_assign_pointer(fn->leaf, rt);
1418 			}
1419 		}
1420 		fn = sn;
1421 	}
1422 #endif
1423 
1424 	err = fib6_add_rt2node(fn, rt, info, extack);
1425 	if (!err) {
1426 		if (rt->nh)
1427 			list_add(&rt->nh_list, &rt->nh->f6i_list);
1428 		__fib6_update_sernum_upto_root(rt, sernum);
1429 		fib6_start_gc(info->nl_net, rt);
1430 	}
1431 
1432 out:
1433 	if (err) {
1434 #ifdef CONFIG_IPV6_SUBTREES
1435 		/*
1436 		 * If fib6_add_1 has cleared the old leaf pointer in the
1437 		 * super-tree leaf node we have to find a new one for it.
1438 		 */
1439 		if (pn != fn) {
1440 			struct fib6_info *pn_leaf =
1441 				rcu_dereference_protected(pn->leaf,
1442 				    lockdep_is_held(&table->tb6_lock));
1443 			if (pn_leaf == rt) {
1444 				pn_leaf = NULL;
1445 				RCU_INIT_POINTER(pn->leaf, NULL);
1446 				fib6_info_release(rt);
1447 			}
1448 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1449 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1450 							   pn);
1451 #if RT6_DEBUG >= 2
1452 				if (!pn_leaf) {
1453 					WARN_ON(!pn_leaf);
1454 					pn_leaf =
1455 					    info->nl_net->ipv6.fib6_null_entry;
1456 				}
1457 #endif
1458 				fib6_info_hold(pn_leaf);
1459 				rcu_assign_pointer(pn->leaf, pn_leaf);
1460 			}
1461 		}
1462 #endif
1463 		goto failure;
1464 	} else if (fib6_requires_src(rt)) {
1465 		fib6_routes_require_src_inc(info->nl_net);
1466 	}
1467 	return err;
1468 
1469 failure:
1470 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1471 	 * 1. fn is an intermediate node and we failed to add the new
1472 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1473 	 * failure case.
1474 	 * 2. fn is the root node in the table and we fail to add the first
1475 	 * default route to it.
1476 	 */
1477 	if (fn &&
1478 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1479 	     (fn->fn_flags & RTN_TL_ROOT &&
1480 	      !rcu_access_pointer(fn->leaf))))
1481 		fib6_repair_tree(info->nl_net, table, fn);
1482 	return err;
1483 }
1484 
1485 /*
1486  *	Routing tree lookup
1487  *
1488  */
1489 
1490 struct lookup_args {
1491 	int			offset;		/* key offset on fib6_info */
1492 	const struct in6_addr	*addr;		/* search key			*/
1493 };
1494 
1495 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1496 					    struct lookup_args *args)
1497 {
1498 	struct fib6_node *fn;
1499 	__be32 dir;
1500 
1501 	if (unlikely(args->offset == 0))
1502 		return NULL;
1503 
1504 	/*
1505 	 *	Descend on a tree
1506 	 */
1507 
1508 	fn = root;
1509 
1510 	for (;;) {
1511 		struct fib6_node *next;
1512 
1513 		dir = addr_bit_set(args->addr, fn->fn_bit);
1514 
1515 		next = dir ? rcu_dereference(fn->right) :
1516 			     rcu_dereference(fn->left);
1517 
1518 		if (next) {
1519 			fn = next;
1520 			continue;
1521 		}
1522 		break;
1523 	}
1524 
1525 	while (fn) {
1526 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1527 
1528 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1529 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1530 			struct rt6key *key;
1531 
1532 			if (!leaf)
1533 				goto backtrack;
1534 
1535 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1536 
1537 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1538 #ifdef CONFIG_IPV6_SUBTREES
1539 				if (subtree) {
1540 					struct fib6_node *sfn;
1541 					sfn = fib6_node_lookup_1(subtree,
1542 								 args + 1);
1543 					if (!sfn)
1544 						goto backtrack;
1545 					fn = sfn;
1546 				}
1547 #endif
1548 				if (fn->fn_flags & RTN_RTINFO)
1549 					return fn;
1550 			}
1551 		}
1552 backtrack:
1553 		if (fn->fn_flags & RTN_ROOT)
1554 			break;
1555 
1556 		fn = rcu_dereference(fn->parent);
1557 	}
1558 
1559 	return NULL;
1560 }
1561 
1562 /* called with rcu_read_lock() held
1563  */
1564 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1565 				   const struct in6_addr *daddr,
1566 				   const struct in6_addr *saddr)
1567 {
1568 	struct fib6_node *fn;
1569 	struct lookup_args args[] = {
1570 		{
1571 			.offset = offsetof(struct fib6_info, fib6_dst),
1572 			.addr = daddr,
1573 		},
1574 #ifdef CONFIG_IPV6_SUBTREES
1575 		{
1576 			.offset = offsetof(struct fib6_info, fib6_src),
1577 			.addr = saddr,
1578 		},
1579 #endif
1580 		{
1581 			.offset = 0,	/* sentinel */
1582 		}
1583 	};
1584 
1585 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1586 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1587 		fn = root;
1588 
1589 	return fn;
1590 }
1591 
1592 /*
1593  *	Get node with specified destination prefix (and source prefix,
1594  *	if subtrees are used)
1595  *	exact_match == true means we try to find fn with exact match of
1596  *	the passed in prefix addr
1597  *	exact_match == false means we try to find fn with longest prefix
1598  *	match of the passed in prefix addr. This is useful for finding fn
1599  *	for cached route as it will be stored in the exception table under
1600  *	the node with longest prefix length.
1601  */
1602 
1603 
1604 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1605 				       const struct in6_addr *addr,
1606 				       int plen, int offset,
1607 				       bool exact_match)
1608 {
1609 	struct fib6_node *fn, *prev = NULL;
1610 
1611 	for (fn = root; fn ; ) {
1612 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1613 		struct rt6key *key;
1614 
1615 		/* This node is being deleted */
1616 		if (!leaf) {
1617 			if (plen <= fn->fn_bit)
1618 				goto out;
1619 			else
1620 				goto next;
1621 		}
1622 
1623 		key = (struct rt6key *)((u8 *)leaf + offset);
1624 
1625 		/*
1626 		 *	Prefix match
1627 		 */
1628 		if (plen < fn->fn_bit ||
1629 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1630 			goto out;
1631 
1632 		if (plen == fn->fn_bit)
1633 			return fn;
1634 
1635 		if (fn->fn_flags & RTN_RTINFO)
1636 			prev = fn;
1637 
1638 next:
1639 		/*
1640 		 *	We have more bits to go
1641 		 */
1642 		if (addr_bit_set(addr, fn->fn_bit))
1643 			fn = rcu_dereference(fn->right);
1644 		else
1645 			fn = rcu_dereference(fn->left);
1646 	}
1647 out:
1648 	if (exact_match)
1649 		return NULL;
1650 	else
1651 		return prev;
1652 }
1653 
1654 struct fib6_node *fib6_locate(struct fib6_node *root,
1655 			      const struct in6_addr *daddr, int dst_len,
1656 			      const struct in6_addr *saddr, int src_len,
1657 			      bool exact_match)
1658 {
1659 	struct fib6_node *fn;
1660 
1661 	fn = fib6_locate_1(root, daddr, dst_len,
1662 			   offsetof(struct fib6_info, fib6_dst),
1663 			   exact_match);
1664 
1665 #ifdef CONFIG_IPV6_SUBTREES
1666 	if (src_len) {
1667 		WARN_ON(saddr == NULL);
1668 		if (fn) {
1669 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1670 
1671 			if (subtree) {
1672 				fn = fib6_locate_1(subtree, saddr, src_len,
1673 					   offsetof(struct fib6_info, fib6_src),
1674 					   exact_match);
1675 			}
1676 		}
1677 	}
1678 #endif
1679 
1680 	if (fn && fn->fn_flags & RTN_RTINFO)
1681 		return fn;
1682 
1683 	return NULL;
1684 }
1685 
1686 
1687 /*
1688  *	Deletion
1689  *
1690  */
1691 
1692 static struct fib6_info *fib6_find_prefix(struct net *net,
1693 					 struct fib6_table *table,
1694 					 struct fib6_node *fn)
1695 {
1696 	struct fib6_node *child_left, *child_right;
1697 
1698 	if (fn->fn_flags & RTN_ROOT)
1699 		return net->ipv6.fib6_null_entry;
1700 
1701 	while (fn) {
1702 		child_left = rcu_dereference_protected(fn->left,
1703 				    lockdep_is_held(&table->tb6_lock));
1704 		child_right = rcu_dereference_protected(fn->right,
1705 				    lockdep_is_held(&table->tb6_lock));
1706 		if (child_left)
1707 			return rcu_dereference_protected(child_left->leaf,
1708 					lockdep_is_held(&table->tb6_lock));
1709 		if (child_right)
1710 			return rcu_dereference_protected(child_right->leaf,
1711 					lockdep_is_held(&table->tb6_lock));
1712 
1713 		fn = FIB6_SUBTREE(fn);
1714 	}
1715 	return NULL;
1716 }
1717 
1718 /*
1719  *	Called to trim the tree of intermediate nodes when possible. "fn"
1720  *	is the node we want to try and remove.
1721  *	Need to own table->tb6_lock
1722  */
1723 
1724 static struct fib6_node *fib6_repair_tree(struct net *net,
1725 					  struct fib6_table *table,
1726 					  struct fib6_node *fn)
1727 {
1728 	int children;
1729 	int nstate;
1730 	struct fib6_node *child;
1731 	struct fib6_walker *w;
1732 	int iter = 0;
1733 
1734 	/* Set fn->leaf to null_entry for root node. */
1735 	if (fn->fn_flags & RTN_TL_ROOT) {
1736 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1737 		return fn;
1738 	}
1739 
1740 	for (;;) {
1741 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1742 					    lockdep_is_held(&table->tb6_lock));
1743 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1744 					    lockdep_is_held(&table->tb6_lock));
1745 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1746 					    lockdep_is_held(&table->tb6_lock));
1747 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1748 					    lockdep_is_held(&table->tb6_lock));
1749 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1750 					    lockdep_is_held(&table->tb6_lock));
1751 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1752 					    lockdep_is_held(&table->tb6_lock));
1753 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1754 					    lockdep_is_held(&table->tb6_lock));
1755 		struct fib6_info *new_fn_leaf;
1756 
1757 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1758 		iter++;
1759 
1760 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1761 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1762 		WARN_ON(fn_leaf);
1763 
1764 		children = 0;
1765 		child = NULL;
1766 		if (fn_r)
1767 			child = fn_r, children |= 1;
1768 		if (fn_l)
1769 			child = fn_l, children |= 2;
1770 
1771 		if (children == 3 || FIB6_SUBTREE(fn)
1772 #ifdef CONFIG_IPV6_SUBTREES
1773 		    /* Subtree root (i.e. fn) may have one child */
1774 		    || (children && fn->fn_flags & RTN_ROOT)
1775 #endif
1776 		    ) {
1777 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1778 #if RT6_DEBUG >= 2
1779 			if (!new_fn_leaf) {
1780 				WARN_ON(!new_fn_leaf);
1781 				new_fn_leaf = net->ipv6.fib6_null_entry;
1782 			}
1783 #endif
1784 			fib6_info_hold(new_fn_leaf);
1785 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1786 			return pn;
1787 		}
1788 
1789 #ifdef CONFIG_IPV6_SUBTREES
1790 		if (FIB6_SUBTREE(pn) == fn) {
1791 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1792 			RCU_INIT_POINTER(pn->subtree, NULL);
1793 			nstate = FWS_L;
1794 		} else {
1795 			WARN_ON(fn->fn_flags & RTN_ROOT);
1796 #endif
1797 			if (pn_r == fn)
1798 				rcu_assign_pointer(pn->right, child);
1799 			else if (pn_l == fn)
1800 				rcu_assign_pointer(pn->left, child);
1801 #if RT6_DEBUG >= 2
1802 			else
1803 				WARN_ON(1);
1804 #endif
1805 			if (child)
1806 				rcu_assign_pointer(child->parent, pn);
1807 			nstate = FWS_R;
1808 #ifdef CONFIG_IPV6_SUBTREES
1809 		}
1810 #endif
1811 
1812 		read_lock(&net->ipv6.fib6_walker_lock);
1813 		FOR_WALKERS(net, w) {
1814 			if (!child) {
1815 				if (w->node == fn) {
1816 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1817 					w->node = pn;
1818 					w->state = nstate;
1819 				}
1820 			} else {
1821 				if (w->node == fn) {
1822 					w->node = child;
1823 					if (children&2) {
1824 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1825 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1826 					} else {
1827 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1828 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1829 					}
1830 				}
1831 			}
1832 		}
1833 		read_unlock(&net->ipv6.fib6_walker_lock);
1834 
1835 		node_free(net, fn);
1836 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1837 			return pn;
1838 
1839 		RCU_INIT_POINTER(pn->leaf, NULL);
1840 		fib6_info_release(pn_leaf);
1841 		fn = pn;
1842 	}
1843 }
1844 
1845 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1846 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1847 {
1848 	struct fib6_walker *w;
1849 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1850 				    lockdep_is_held(&table->tb6_lock));
1851 	struct net *net = info->nl_net;
1852 
1853 	RT6_TRACE("fib6_del_route\n");
1854 
1855 	/* Unlink it */
1856 	*rtp = rt->fib6_next;
1857 	rt->fib6_node = NULL;
1858 	net->ipv6.rt6_stats->fib_rt_entries--;
1859 	net->ipv6.rt6_stats->fib_discarded_routes++;
1860 
1861 	/* Flush all cached dst in exception table */
1862 	rt6_flush_exceptions(rt);
1863 
1864 	/* Reset round-robin state, if necessary */
1865 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1866 		fn->rr_ptr = NULL;
1867 
1868 	/* Remove this entry from other siblings */
1869 	if (rt->fib6_nsiblings) {
1870 		struct fib6_info *sibling, *next_sibling;
1871 
1872 		list_for_each_entry_safe(sibling, next_sibling,
1873 					 &rt->fib6_siblings, fib6_siblings)
1874 			sibling->fib6_nsiblings--;
1875 		rt->fib6_nsiblings = 0;
1876 		list_del_init(&rt->fib6_siblings);
1877 		rt6_multipath_rebalance(next_sibling);
1878 	}
1879 
1880 	/* Adjust walkers */
1881 	read_lock(&net->ipv6.fib6_walker_lock);
1882 	FOR_WALKERS(net, w) {
1883 		if (w->state == FWS_C && w->leaf == rt) {
1884 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1885 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1886 					    lockdep_is_held(&table->tb6_lock));
1887 			if (!w->leaf)
1888 				w->state = FWS_U;
1889 		}
1890 	}
1891 	read_unlock(&net->ipv6.fib6_walker_lock);
1892 
1893 	/* If it was last route, call fib6_repair_tree() to:
1894 	 * 1. For root node, put back null_entry as how the table was created.
1895 	 * 2. For other nodes, expunge its radix tree node.
1896 	 */
1897 	if (!rcu_access_pointer(fn->leaf)) {
1898 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1899 			fn->fn_flags &= ~RTN_RTINFO;
1900 			net->ipv6.rt6_stats->fib_route_nodes--;
1901 		}
1902 		fn = fib6_repair_tree(net, table, fn);
1903 	}
1904 
1905 	fib6_purge_rt(rt, fn, net);
1906 
1907 	if (!info->skip_notify_kernel)
1908 		call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1909 	if (!info->skip_notify)
1910 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1911 
1912 	fib6_info_release(rt);
1913 }
1914 
1915 /* Need to own table->tb6_lock */
1916 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1917 {
1918 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1919 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1920 	struct fib6_table *table = rt->fib6_table;
1921 	struct net *net = info->nl_net;
1922 	struct fib6_info __rcu **rtp;
1923 	struct fib6_info __rcu **rtp_next;
1924 
1925 	if (!fn || rt == net->ipv6.fib6_null_entry)
1926 		return -ENOENT;
1927 
1928 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1929 
1930 	/*
1931 	 *	Walk the leaf entries looking for ourself
1932 	 */
1933 
1934 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1935 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1936 					lockdep_is_held(&table->tb6_lock));
1937 		if (rt == cur) {
1938 			if (fib6_requires_src(cur))
1939 				fib6_routes_require_src_dec(info->nl_net);
1940 			fib6_del_route(table, fn, rtp, info);
1941 			return 0;
1942 		}
1943 		rtp_next = &cur->fib6_next;
1944 	}
1945 	return -ENOENT;
1946 }
1947 
1948 /*
1949  *	Tree traversal function.
1950  *
1951  *	Certainly, it is not interrupt safe.
1952  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1953  *	It means, that we can modify tree during walking
1954  *	and use this function for garbage collection, clone pruning,
1955  *	cleaning tree when a device goes down etc. etc.
1956  *
1957  *	It guarantees that every node will be traversed,
1958  *	and that it will be traversed only once.
1959  *
1960  *	Callback function w->func may return:
1961  *	0 -> continue walking.
1962  *	positive value -> walking is suspended (used by tree dumps,
1963  *	and probably by gc, if it will be split to several slices)
1964  *	negative value -> terminate walking.
1965  *
1966  *	The function itself returns:
1967  *	0   -> walk is complete.
1968  *	>0  -> walk is incomplete (i.e. suspended)
1969  *	<0  -> walk is terminated by an error.
1970  *
1971  *	This function is called with tb6_lock held.
1972  */
1973 
1974 static int fib6_walk_continue(struct fib6_walker *w)
1975 {
1976 	struct fib6_node *fn, *pn, *left, *right;
1977 
1978 	/* w->root should always be table->tb6_root */
1979 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1980 
1981 	for (;;) {
1982 		fn = w->node;
1983 		if (!fn)
1984 			return 0;
1985 
1986 		switch (w->state) {
1987 #ifdef CONFIG_IPV6_SUBTREES
1988 		case FWS_S:
1989 			if (FIB6_SUBTREE(fn)) {
1990 				w->node = FIB6_SUBTREE(fn);
1991 				continue;
1992 			}
1993 			w->state = FWS_L;
1994 #endif
1995 			/* fall through */
1996 		case FWS_L:
1997 			left = rcu_dereference_protected(fn->left, 1);
1998 			if (left) {
1999 				w->node = left;
2000 				w->state = FWS_INIT;
2001 				continue;
2002 			}
2003 			w->state = FWS_R;
2004 			/* fall through */
2005 		case FWS_R:
2006 			right = rcu_dereference_protected(fn->right, 1);
2007 			if (right) {
2008 				w->node = right;
2009 				w->state = FWS_INIT;
2010 				continue;
2011 			}
2012 			w->state = FWS_C;
2013 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
2014 			/* fall through */
2015 		case FWS_C:
2016 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2017 				int err;
2018 
2019 				if (w->skip) {
2020 					w->skip--;
2021 					goto skip;
2022 				}
2023 
2024 				err = w->func(w);
2025 				if (err)
2026 					return err;
2027 
2028 				w->count++;
2029 				continue;
2030 			}
2031 skip:
2032 			w->state = FWS_U;
2033 			/* fall through */
2034 		case FWS_U:
2035 			if (fn == w->root)
2036 				return 0;
2037 			pn = rcu_dereference_protected(fn->parent, 1);
2038 			left = rcu_dereference_protected(pn->left, 1);
2039 			right = rcu_dereference_protected(pn->right, 1);
2040 			w->node = pn;
2041 #ifdef CONFIG_IPV6_SUBTREES
2042 			if (FIB6_SUBTREE(pn) == fn) {
2043 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
2044 				w->state = FWS_L;
2045 				continue;
2046 			}
2047 #endif
2048 			if (left == fn) {
2049 				w->state = FWS_R;
2050 				continue;
2051 			}
2052 			if (right == fn) {
2053 				w->state = FWS_C;
2054 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2055 				continue;
2056 			}
2057 #if RT6_DEBUG >= 2
2058 			WARN_ON(1);
2059 #endif
2060 		}
2061 	}
2062 }
2063 
2064 static int fib6_walk(struct net *net, struct fib6_walker *w)
2065 {
2066 	int res;
2067 
2068 	w->state = FWS_INIT;
2069 	w->node = w->root;
2070 
2071 	fib6_walker_link(net, w);
2072 	res = fib6_walk_continue(w);
2073 	if (res <= 0)
2074 		fib6_walker_unlink(net, w);
2075 	return res;
2076 }
2077 
2078 static int fib6_clean_node(struct fib6_walker *w)
2079 {
2080 	int res;
2081 	struct fib6_info *rt;
2082 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2083 	struct nl_info info = {
2084 		.nl_net = c->net,
2085 		.skip_notify = c->skip_notify,
2086 	};
2087 
2088 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2089 	    w->node->fn_sernum != c->sernum)
2090 		w->node->fn_sernum = c->sernum;
2091 
2092 	if (!c->func) {
2093 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2094 		w->leaf = NULL;
2095 		return 0;
2096 	}
2097 
2098 	for_each_fib6_walker_rt(w) {
2099 		res = c->func(rt, c->arg);
2100 		if (res == -1) {
2101 			w->leaf = rt;
2102 			res = fib6_del(rt, &info);
2103 			if (res) {
2104 #if RT6_DEBUG >= 2
2105 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2106 					 __func__, rt,
2107 					 rcu_access_pointer(rt->fib6_node),
2108 					 res);
2109 #endif
2110 				continue;
2111 			}
2112 			return 0;
2113 		} else if (res == -2) {
2114 			if (WARN_ON(!rt->fib6_nsiblings))
2115 				continue;
2116 			rt = list_last_entry(&rt->fib6_siblings,
2117 					     struct fib6_info, fib6_siblings);
2118 			continue;
2119 		}
2120 		WARN_ON(res != 0);
2121 	}
2122 	w->leaf = rt;
2123 	return 0;
2124 }
2125 
2126 /*
2127  *	Convenient frontend to tree walker.
2128  *
2129  *	func is called on each route.
2130  *		It may return -2 -> skip multipath route.
2131  *			      -1 -> delete this route.
2132  *		              0  -> continue walking
2133  */
2134 
2135 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2136 			    int (*func)(struct fib6_info *, void *arg),
2137 			    int sernum, void *arg, bool skip_notify)
2138 {
2139 	struct fib6_cleaner c;
2140 
2141 	c.w.root = root;
2142 	c.w.func = fib6_clean_node;
2143 	c.w.count = 0;
2144 	c.w.skip = 0;
2145 	c.w.skip_in_node = 0;
2146 	c.func = func;
2147 	c.sernum = sernum;
2148 	c.arg = arg;
2149 	c.net = net;
2150 	c.skip_notify = skip_notify;
2151 
2152 	fib6_walk(net, &c.w);
2153 }
2154 
2155 static void __fib6_clean_all(struct net *net,
2156 			     int (*func)(struct fib6_info *, void *),
2157 			     int sernum, void *arg, bool skip_notify)
2158 {
2159 	struct fib6_table *table;
2160 	struct hlist_head *head;
2161 	unsigned int h;
2162 
2163 	rcu_read_lock();
2164 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2165 		head = &net->ipv6.fib_table_hash[h];
2166 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2167 			spin_lock_bh(&table->tb6_lock);
2168 			fib6_clean_tree(net, &table->tb6_root,
2169 					func, sernum, arg, skip_notify);
2170 			spin_unlock_bh(&table->tb6_lock);
2171 		}
2172 	}
2173 	rcu_read_unlock();
2174 }
2175 
2176 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2177 		    void *arg)
2178 {
2179 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2180 }
2181 
2182 void fib6_clean_all_skip_notify(struct net *net,
2183 				int (*func)(struct fib6_info *, void *),
2184 				void *arg)
2185 {
2186 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2187 }
2188 
2189 static void fib6_flush_trees(struct net *net)
2190 {
2191 	int new_sernum = fib6_new_sernum(net);
2192 
2193 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2194 }
2195 
2196 /*
2197  *	Garbage collection
2198  */
2199 
2200 static int fib6_age(struct fib6_info *rt, void *arg)
2201 {
2202 	struct fib6_gc_args *gc_args = arg;
2203 	unsigned long now = jiffies;
2204 
2205 	/*
2206 	 *	check addrconf expiration here.
2207 	 *	Routes are expired even if they are in use.
2208 	 */
2209 
2210 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2211 		if (time_after(now, rt->expires)) {
2212 			RT6_TRACE("expiring %p\n", rt);
2213 			return -1;
2214 		}
2215 		gc_args->more++;
2216 	}
2217 
2218 	/*	Also age clones in the exception table.
2219 	 *	Note, that clones are aged out
2220 	 *	only if they are not in use now.
2221 	 */
2222 	rt6_age_exceptions(rt, gc_args, now);
2223 
2224 	return 0;
2225 }
2226 
2227 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2228 {
2229 	struct fib6_gc_args gc_args;
2230 	unsigned long now;
2231 
2232 	if (force) {
2233 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2234 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2235 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2236 		return;
2237 	}
2238 	gc_args.timeout = expires ? (int)expires :
2239 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2240 	gc_args.more = 0;
2241 
2242 	fib6_clean_all(net, fib6_age, &gc_args);
2243 	now = jiffies;
2244 	net->ipv6.ip6_rt_last_gc = now;
2245 
2246 	if (gc_args.more)
2247 		mod_timer(&net->ipv6.ip6_fib_timer,
2248 			  round_jiffies(now
2249 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2250 	else
2251 		del_timer(&net->ipv6.ip6_fib_timer);
2252 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2253 }
2254 
2255 static void fib6_gc_timer_cb(struct timer_list *t)
2256 {
2257 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2258 
2259 	fib6_run_gc(0, arg, true);
2260 }
2261 
2262 static int __net_init fib6_net_init(struct net *net)
2263 {
2264 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2265 	int err;
2266 
2267 	err = fib6_notifier_init(net);
2268 	if (err)
2269 		return err;
2270 
2271 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2272 	rwlock_init(&net->ipv6.fib6_walker_lock);
2273 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2274 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2275 
2276 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2277 	if (!net->ipv6.rt6_stats)
2278 		goto out_timer;
2279 
2280 	/* Avoid false sharing : Use at least a full cache line */
2281 	size = max_t(size_t, size, L1_CACHE_BYTES);
2282 
2283 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2284 	if (!net->ipv6.fib_table_hash)
2285 		goto out_rt6_stats;
2286 
2287 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2288 					  GFP_KERNEL);
2289 	if (!net->ipv6.fib6_main_tbl)
2290 		goto out_fib_table_hash;
2291 
2292 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2293 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2294 			   net->ipv6.fib6_null_entry);
2295 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2296 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2297 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2298 
2299 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2300 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2301 					   GFP_KERNEL);
2302 	if (!net->ipv6.fib6_local_tbl)
2303 		goto out_fib6_main_tbl;
2304 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2305 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2306 			   net->ipv6.fib6_null_entry);
2307 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2308 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2309 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2310 #endif
2311 	fib6_tables_init(net);
2312 
2313 	return 0;
2314 
2315 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2316 out_fib6_main_tbl:
2317 	kfree(net->ipv6.fib6_main_tbl);
2318 #endif
2319 out_fib_table_hash:
2320 	kfree(net->ipv6.fib_table_hash);
2321 out_rt6_stats:
2322 	kfree(net->ipv6.rt6_stats);
2323 out_timer:
2324 	fib6_notifier_exit(net);
2325 	return -ENOMEM;
2326 }
2327 
2328 static void fib6_net_exit(struct net *net)
2329 {
2330 	unsigned int i;
2331 
2332 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2333 
2334 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2335 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2336 		struct hlist_node *tmp;
2337 		struct fib6_table *tb;
2338 
2339 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2340 			hlist_del(&tb->tb6_hlist);
2341 			fib6_free_table(tb);
2342 		}
2343 	}
2344 
2345 	kfree(net->ipv6.fib_table_hash);
2346 	kfree(net->ipv6.rt6_stats);
2347 	fib6_notifier_exit(net);
2348 }
2349 
2350 static struct pernet_operations fib6_net_ops = {
2351 	.init = fib6_net_init,
2352 	.exit = fib6_net_exit,
2353 };
2354 
2355 int __init fib6_init(void)
2356 {
2357 	int ret = -ENOMEM;
2358 
2359 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2360 					   sizeof(struct fib6_node),
2361 					   0, SLAB_HWCACHE_ALIGN,
2362 					   NULL);
2363 	if (!fib6_node_kmem)
2364 		goto out;
2365 
2366 	ret = register_pernet_subsys(&fib6_net_ops);
2367 	if (ret)
2368 		goto out_kmem_cache_create;
2369 
2370 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2371 				   inet6_dump_fib, 0);
2372 	if (ret)
2373 		goto out_unregister_subsys;
2374 
2375 	__fib6_flush_trees = fib6_flush_trees;
2376 out:
2377 	return ret;
2378 
2379 out_unregister_subsys:
2380 	unregister_pernet_subsys(&fib6_net_ops);
2381 out_kmem_cache_create:
2382 	kmem_cache_destroy(fib6_node_kmem);
2383 	goto out;
2384 }
2385 
2386 void fib6_gc_cleanup(void)
2387 {
2388 	unregister_pernet_subsys(&fib6_net_ops);
2389 	kmem_cache_destroy(fib6_node_kmem);
2390 }
2391 
2392 #ifdef CONFIG_PROC_FS
2393 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2394 {
2395 	struct fib6_info *rt = v;
2396 	struct ipv6_route_iter *iter = seq->private;
2397 	struct fib6_nh *fib6_nh = rt->fib6_nh;
2398 	unsigned int flags = rt->fib6_flags;
2399 	const struct net_device *dev;
2400 
2401 	if (rt->nh)
2402 		fib6_nh = nexthop_fib6_nh(rt->nh);
2403 
2404 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2405 
2406 #ifdef CONFIG_IPV6_SUBTREES
2407 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2408 #else
2409 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2410 #endif
2411 	if (fib6_nh->fib_nh_gw_family) {
2412 		flags |= RTF_GATEWAY;
2413 		seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2414 	} else {
2415 		seq_puts(seq, "00000000000000000000000000000000");
2416 	}
2417 
2418 	dev = fib6_nh->fib_nh_dev;
2419 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2420 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2421 		   flags, dev ? dev->name : "");
2422 	iter->w.leaf = NULL;
2423 	return 0;
2424 }
2425 
2426 static int ipv6_route_yield(struct fib6_walker *w)
2427 {
2428 	struct ipv6_route_iter *iter = w->args;
2429 
2430 	if (!iter->skip)
2431 		return 1;
2432 
2433 	do {
2434 		iter->w.leaf = rcu_dereference_protected(
2435 				iter->w.leaf->fib6_next,
2436 				lockdep_is_held(&iter->tbl->tb6_lock));
2437 		iter->skip--;
2438 		if (!iter->skip && iter->w.leaf)
2439 			return 1;
2440 	} while (iter->w.leaf);
2441 
2442 	return 0;
2443 }
2444 
2445 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2446 				      struct net *net)
2447 {
2448 	memset(&iter->w, 0, sizeof(iter->w));
2449 	iter->w.func = ipv6_route_yield;
2450 	iter->w.root = &iter->tbl->tb6_root;
2451 	iter->w.state = FWS_INIT;
2452 	iter->w.node = iter->w.root;
2453 	iter->w.args = iter;
2454 	iter->sernum = iter->w.root->fn_sernum;
2455 	INIT_LIST_HEAD(&iter->w.lh);
2456 	fib6_walker_link(net, &iter->w);
2457 }
2458 
2459 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2460 						    struct net *net)
2461 {
2462 	unsigned int h;
2463 	struct hlist_node *node;
2464 
2465 	if (tbl) {
2466 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2467 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2468 	} else {
2469 		h = 0;
2470 		node = NULL;
2471 	}
2472 
2473 	while (!node && h < FIB6_TABLE_HASHSZ) {
2474 		node = rcu_dereference_bh(
2475 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2476 	}
2477 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2478 }
2479 
2480 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2481 {
2482 	if (iter->sernum != iter->w.root->fn_sernum) {
2483 		iter->sernum = iter->w.root->fn_sernum;
2484 		iter->w.state = FWS_INIT;
2485 		iter->w.node = iter->w.root;
2486 		WARN_ON(iter->w.skip);
2487 		iter->w.skip = iter->w.count;
2488 	}
2489 }
2490 
2491 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2492 {
2493 	int r;
2494 	struct fib6_info *n;
2495 	struct net *net = seq_file_net(seq);
2496 	struct ipv6_route_iter *iter = seq->private;
2497 
2498 	if (!v)
2499 		goto iter_table;
2500 
2501 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2502 	if (n) {
2503 		++*pos;
2504 		return n;
2505 	}
2506 
2507 iter_table:
2508 	ipv6_route_check_sernum(iter);
2509 	spin_lock_bh(&iter->tbl->tb6_lock);
2510 	r = fib6_walk_continue(&iter->w);
2511 	spin_unlock_bh(&iter->tbl->tb6_lock);
2512 	if (r > 0) {
2513 		if (v)
2514 			++*pos;
2515 		return iter->w.leaf;
2516 	} else if (r < 0) {
2517 		fib6_walker_unlink(net, &iter->w);
2518 		return NULL;
2519 	}
2520 	fib6_walker_unlink(net, &iter->w);
2521 
2522 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2523 	if (!iter->tbl)
2524 		return NULL;
2525 
2526 	ipv6_route_seq_setup_walk(iter, net);
2527 	goto iter_table;
2528 }
2529 
2530 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2531 	__acquires(RCU_BH)
2532 {
2533 	struct net *net = seq_file_net(seq);
2534 	struct ipv6_route_iter *iter = seq->private;
2535 
2536 	rcu_read_lock_bh();
2537 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2538 	iter->skip = *pos;
2539 
2540 	if (iter->tbl) {
2541 		ipv6_route_seq_setup_walk(iter, net);
2542 		return ipv6_route_seq_next(seq, NULL, pos);
2543 	} else {
2544 		return NULL;
2545 	}
2546 }
2547 
2548 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2549 {
2550 	struct fib6_walker *w = &iter->w;
2551 	return w->node && !(w->state == FWS_U && w->node == w->root);
2552 }
2553 
2554 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2555 	__releases(RCU_BH)
2556 {
2557 	struct net *net = seq_file_net(seq);
2558 	struct ipv6_route_iter *iter = seq->private;
2559 
2560 	if (ipv6_route_iter_active(iter))
2561 		fib6_walker_unlink(net, &iter->w);
2562 
2563 	rcu_read_unlock_bh();
2564 }
2565 
2566 const struct seq_operations ipv6_route_seq_ops = {
2567 	.start	= ipv6_route_seq_start,
2568 	.next	= ipv6_route_seq_next,
2569 	.stop	= ipv6_route_seq_stop,
2570 	.show	= ipv6_route_seq_show
2571 };
2572 #endif /* CONFIG_PROC_FS */
2573