xref: /openbmc/linux/net/ipv6/ip6_fib.c (revision 675aaf05)
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, w->skip_in_node);
468 		if (res >= 0) {
469 			/* Frame is full, suspend walking */
470 			w->leaf = rt;
471 
472 			/* We'll restart from this node, so if some routes were
473 			 * already dumped, skip them next time.
474 			 */
475 			w->skip_in_node += res;
476 
477 			return 1;
478 		}
479 		w->skip_in_node = 0;
480 
481 		/* Multipath routes are dumped in one route with the
482 		 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
483 		 * last sibling of this route (no need to dump the
484 		 * sibling routes again)
485 		 */
486 		if (rt->fib6_nsiblings)
487 			rt = list_last_entry(&rt->fib6_siblings,
488 					     struct fib6_info,
489 					     fib6_siblings);
490 	}
491 	w->leaf = NULL;
492 	return 0;
493 }
494 
495 static void fib6_dump_end(struct netlink_callback *cb)
496 {
497 	struct net *net = sock_net(cb->skb->sk);
498 	struct fib6_walker *w = (void *)cb->args[2];
499 
500 	if (w) {
501 		if (cb->args[4]) {
502 			cb->args[4] = 0;
503 			fib6_walker_unlink(net, w);
504 		}
505 		cb->args[2] = 0;
506 		kfree(w);
507 	}
508 	cb->done = (void *)cb->args[3];
509 	cb->args[1] = 3;
510 }
511 
512 static int fib6_dump_done(struct netlink_callback *cb)
513 {
514 	fib6_dump_end(cb);
515 	return cb->done ? cb->done(cb) : 0;
516 }
517 
518 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
519 			   struct netlink_callback *cb)
520 {
521 	struct net *net = sock_net(skb->sk);
522 	struct fib6_walker *w;
523 	int res;
524 
525 	w = (void *)cb->args[2];
526 	w->root = &table->tb6_root;
527 
528 	if (cb->args[4] == 0) {
529 		w->count = 0;
530 		w->skip = 0;
531 		w->skip_in_node = 0;
532 
533 		spin_lock_bh(&table->tb6_lock);
534 		res = fib6_walk(net, w);
535 		spin_unlock_bh(&table->tb6_lock);
536 		if (res > 0) {
537 			cb->args[4] = 1;
538 			cb->args[5] = w->root->fn_sernum;
539 		}
540 	} else {
541 		if (cb->args[5] != w->root->fn_sernum) {
542 			/* Begin at the root if the tree changed */
543 			cb->args[5] = w->root->fn_sernum;
544 			w->state = FWS_INIT;
545 			w->node = w->root;
546 			w->skip = w->count;
547 			w->skip_in_node = 0;
548 		} else
549 			w->skip = 0;
550 
551 		spin_lock_bh(&table->tb6_lock);
552 		res = fib6_walk_continue(w);
553 		spin_unlock_bh(&table->tb6_lock);
554 		if (res <= 0) {
555 			fib6_walker_unlink(net, w);
556 			cb->args[4] = 0;
557 		}
558 	}
559 
560 	return res;
561 }
562 
563 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
564 {
565 	struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
566 					 .filter.dump_routes = true };
567 	const struct nlmsghdr *nlh = cb->nlh;
568 	struct net *net = sock_net(skb->sk);
569 	unsigned int h, s_h;
570 	unsigned int e = 0, s_e;
571 	struct fib6_walker *w;
572 	struct fib6_table *tb;
573 	struct hlist_head *head;
574 	int res = 0;
575 
576 	if (cb->strict_check) {
577 		int err;
578 
579 		err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
580 		if (err < 0)
581 			return err;
582 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
583 		struct rtmsg *rtm = nlmsg_data(nlh);
584 
585 		if (rtm->rtm_flags & RTM_F_PREFIX)
586 			arg.filter.flags = RTM_F_PREFIX;
587 	}
588 
589 	w = (void *)cb->args[2];
590 	if (!w) {
591 		/* New dump:
592 		 *
593 		 * 1. hook callback destructor.
594 		 */
595 		cb->args[3] = (long)cb->done;
596 		cb->done = fib6_dump_done;
597 
598 		/*
599 		 * 2. allocate and initialize walker.
600 		 */
601 		w = kzalloc(sizeof(*w), GFP_ATOMIC);
602 		if (!w)
603 			return -ENOMEM;
604 		w->func = fib6_dump_node;
605 		cb->args[2] = (long)w;
606 	}
607 
608 	arg.skb = skb;
609 	arg.cb = cb;
610 	arg.net = net;
611 	w->args = &arg;
612 
613 	if (arg.filter.table_id) {
614 		tb = fib6_get_table(net, arg.filter.table_id);
615 		if (!tb) {
616 			if (arg.filter.dump_all_families)
617 				goto out;
618 
619 			NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
620 			return -ENOENT;
621 		}
622 
623 		if (!cb->args[0]) {
624 			res = fib6_dump_table(tb, skb, cb);
625 			if (!res)
626 				cb->args[0] = 1;
627 		}
628 		goto out;
629 	}
630 
631 	s_h = cb->args[0];
632 	s_e = cb->args[1];
633 
634 	rcu_read_lock();
635 	for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
636 		e = 0;
637 		head = &net->ipv6.fib_table_hash[h];
638 		hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
639 			if (e < s_e)
640 				goto next;
641 			res = fib6_dump_table(tb, skb, cb);
642 			if (res != 0)
643 				goto out_unlock;
644 next:
645 			e++;
646 		}
647 	}
648 out_unlock:
649 	rcu_read_unlock();
650 	cb->args[1] = e;
651 	cb->args[0] = h;
652 out:
653 	res = res < 0 ? res : skb->len;
654 	if (res <= 0)
655 		fib6_dump_end(cb);
656 	return res;
657 }
658 
659 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
660 {
661 	if (!f6i)
662 		return;
663 
664 	if (f6i->fib6_metrics == &dst_default_metrics) {
665 		struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
666 
667 		if (!p)
668 			return;
669 
670 		refcount_set(&p->refcnt, 1);
671 		f6i->fib6_metrics = p;
672 	}
673 
674 	f6i->fib6_metrics->metrics[metric - 1] = val;
675 }
676 
677 /*
678  *	Routing Table
679  *
680  *	return the appropriate node for a routing tree "add" operation
681  *	by either creating and inserting or by returning an existing
682  *	node.
683  */
684 
685 static struct fib6_node *fib6_add_1(struct net *net,
686 				    struct fib6_table *table,
687 				    struct fib6_node *root,
688 				    struct in6_addr *addr, int plen,
689 				    int offset, int allow_create,
690 				    int replace_required,
691 				    struct netlink_ext_ack *extack)
692 {
693 	struct fib6_node *fn, *in, *ln;
694 	struct fib6_node *pn = NULL;
695 	struct rt6key *key;
696 	int	bit;
697 	__be32	dir = 0;
698 
699 	RT6_TRACE("fib6_add_1\n");
700 
701 	/* insert node in tree */
702 
703 	fn = root;
704 
705 	do {
706 		struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
707 					    lockdep_is_held(&table->tb6_lock));
708 		key = (struct rt6key *)((u8 *)leaf + offset);
709 
710 		/*
711 		 *	Prefix match
712 		 */
713 		if (plen < fn->fn_bit ||
714 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
715 			if (!allow_create) {
716 				if (replace_required) {
717 					NL_SET_ERR_MSG(extack,
718 						       "Can not replace route - no match found");
719 					pr_warn("Can't replace route, no match found\n");
720 					return ERR_PTR(-ENOENT);
721 				}
722 				pr_warn("NLM_F_CREATE should be set when creating new route\n");
723 			}
724 			goto insert_above;
725 		}
726 
727 		/*
728 		 *	Exact match ?
729 		 */
730 
731 		if (plen == fn->fn_bit) {
732 			/* clean up an intermediate node */
733 			if (!(fn->fn_flags & RTN_RTINFO)) {
734 				RCU_INIT_POINTER(fn->leaf, NULL);
735 				fib6_info_release(leaf);
736 			/* remove null_entry in the root node */
737 			} else if (fn->fn_flags & RTN_TL_ROOT &&
738 				   rcu_access_pointer(fn->leaf) ==
739 				   net->ipv6.fib6_null_entry) {
740 				RCU_INIT_POINTER(fn->leaf, NULL);
741 			}
742 
743 			return fn;
744 		}
745 
746 		/*
747 		 *	We have more bits to go
748 		 */
749 
750 		/* Try to walk down on tree. */
751 		dir = addr_bit_set(addr, fn->fn_bit);
752 		pn = fn;
753 		fn = dir ?
754 		     rcu_dereference_protected(fn->right,
755 					lockdep_is_held(&table->tb6_lock)) :
756 		     rcu_dereference_protected(fn->left,
757 					lockdep_is_held(&table->tb6_lock));
758 	} while (fn);
759 
760 	if (!allow_create) {
761 		/* We should not create new node because
762 		 * NLM_F_REPLACE was specified without NLM_F_CREATE
763 		 * I assume it is safe to require NLM_F_CREATE when
764 		 * REPLACE flag is used! Later we may want to remove the
765 		 * check for replace_required, because according
766 		 * to netlink specification, NLM_F_CREATE
767 		 * MUST be specified if new route is created.
768 		 * That would keep IPv6 consistent with IPv4
769 		 */
770 		if (replace_required) {
771 			NL_SET_ERR_MSG(extack,
772 				       "Can not replace route - no match found");
773 			pr_warn("Can't replace route, no match found\n");
774 			return ERR_PTR(-ENOENT);
775 		}
776 		pr_warn("NLM_F_CREATE should be set when creating new route\n");
777 	}
778 	/*
779 	 *	We walked to the bottom of tree.
780 	 *	Create new leaf node without children.
781 	 */
782 
783 	ln = node_alloc(net);
784 
785 	if (!ln)
786 		return ERR_PTR(-ENOMEM);
787 	ln->fn_bit = plen;
788 	RCU_INIT_POINTER(ln->parent, pn);
789 
790 	if (dir)
791 		rcu_assign_pointer(pn->right, ln);
792 	else
793 		rcu_assign_pointer(pn->left, ln);
794 
795 	return ln;
796 
797 
798 insert_above:
799 	/*
800 	 * split since we don't have a common prefix anymore or
801 	 * we have a less significant route.
802 	 * we've to insert an intermediate node on the list
803 	 * this new node will point to the one we need to create
804 	 * and the current
805 	 */
806 
807 	pn = rcu_dereference_protected(fn->parent,
808 				       lockdep_is_held(&table->tb6_lock));
809 
810 	/* find 1st bit in difference between the 2 addrs.
811 
812 	   See comment in __ipv6_addr_diff: bit may be an invalid value,
813 	   but if it is >= plen, the value is ignored in any case.
814 	 */
815 
816 	bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
817 
818 	/*
819 	 *		(intermediate)[in]
820 	 *	          /	   \
821 	 *	(new leaf node)[ln] (old node)[fn]
822 	 */
823 	if (plen > bit) {
824 		in = node_alloc(net);
825 		ln = node_alloc(net);
826 
827 		if (!in || !ln) {
828 			if (in)
829 				node_free_immediate(net, in);
830 			if (ln)
831 				node_free_immediate(net, ln);
832 			return ERR_PTR(-ENOMEM);
833 		}
834 
835 		/*
836 		 * new intermediate node.
837 		 * RTN_RTINFO will
838 		 * be off since that an address that chooses one of
839 		 * the branches would not match less specific routes
840 		 * in the other branch
841 		 */
842 
843 		in->fn_bit = bit;
844 
845 		RCU_INIT_POINTER(in->parent, pn);
846 		in->leaf = fn->leaf;
847 		fib6_info_hold(rcu_dereference_protected(in->leaf,
848 				lockdep_is_held(&table->tb6_lock)));
849 
850 		/* update parent pointer */
851 		if (dir)
852 			rcu_assign_pointer(pn->right, in);
853 		else
854 			rcu_assign_pointer(pn->left, in);
855 
856 		ln->fn_bit = plen;
857 
858 		RCU_INIT_POINTER(ln->parent, in);
859 		rcu_assign_pointer(fn->parent, in);
860 
861 		if (addr_bit_set(addr, bit)) {
862 			rcu_assign_pointer(in->right, ln);
863 			rcu_assign_pointer(in->left, fn);
864 		} else {
865 			rcu_assign_pointer(in->left, ln);
866 			rcu_assign_pointer(in->right, fn);
867 		}
868 	} else { /* plen <= bit */
869 
870 		/*
871 		 *		(new leaf node)[ln]
872 		 *	          /	   \
873 		 *	     (old node)[fn] NULL
874 		 */
875 
876 		ln = node_alloc(net);
877 
878 		if (!ln)
879 			return ERR_PTR(-ENOMEM);
880 
881 		ln->fn_bit = plen;
882 
883 		RCU_INIT_POINTER(ln->parent, pn);
884 
885 		if (addr_bit_set(&key->addr, plen))
886 			RCU_INIT_POINTER(ln->right, fn);
887 		else
888 			RCU_INIT_POINTER(ln->left, fn);
889 
890 		rcu_assign_pointer(fn->parent, ln);
891 
892 		if (dir)
893 			rcu_assign_pointer(pn->right, ln);
894 		else
895 			rcu_assign_pointer(pn->left, ln);
896 	}
897 	return ln;
898 }
899 
900 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
901 				  const struct fib6_info *match,
902 				  const struct fib6_table *table)
903 {
904 	int cpu;
905 
906 	if (!fib6_nh->rt6i_pcpu)
907 		return;
908 
909 	/* release the reference to this fib entry from
910 	 * all of its cached pcpu routes
911 	 */
912 	for_each_possible_cpu(cpu) {
913 		struct rt6_info **ppcpu_rt;
914 		struct rt6_info *pcpu_rt;
915 
916 		ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
917 		pcpu_rt = *ppcpu_rt;
918 
919 		/* only dropping the 'from' reference if the cached route
920 		 * is using 'match'. The cached pcpu_rt->from only changes
921 		 * from a fib6_info to NULL (ip6_dst_destroy); it can never
922 		 * change from one fib6_info reference to another
923 		 */
924 		if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
925 			struct fib6_info *from;
926 
927 			from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
928 			fib6_info_release(from);
929 		}
930 	}
931 }
932 
933 struct fib6_nh_pcpu_arg {
934 	struct fib6_info	*from;
935 	const struct fib6_table *table;
936 };
937 
938 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
939 {
940 	struct fib6_nh_pcpu_arg *arg = _arg;
941 
942 	__fib6_drop_pcpu_from(nh, arg->from, arg->table);
943 	return 0;
944 }
945 
946 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
947 				const struct fib6_table *table)
948 {
949 	/* Make sure rt6_make_pcpu_route() wont add other percpu routes
950 	 * while we are cleaning them here.
951 	 */
952 	f6i->fib6_destroying = 1;
953 	mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
954 
955 	if (f6i->nh) {
956 		struct fib6_nh_pcpu_arg arg = {
957 			.from = f6i,
958 			.table = table
959 		};
960 
961 		nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
962 					 &arg);
963 	} else {
964 		struct fib6_nh *fib6_nh;
965 
966 		fib6_nh = f6i->fib6_nh;
967 		__fib6_drop_pcpu_from(fib6_nh, f6i, table);
968 	}
969 }
970 
971 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
972 			  struct net *net)
973 {
974 	struct fib6_table *table = rt->fib6_table;
975 
976 	fib6_drop_pcpu_from(rt, table);
977 
978 	if (rt->nh && !list_empty(&rt->nh_list))
979 		list_del_init(&rt->nh_list);
980 
981 	if (refcount_read(&rt->fib6_ref) != 1) {
982 		/* This route is used as dummy address holder in some split
983 		 * nodes. It is not leaked, but it still holds other resources,
984 		 * which must be released in time. So, scan ascendant nodes
985 		 * and replace dummy references to this route with references
986 		 * to still alive ones.
987 		 */
988 		while (fn) {
989 			struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
990 					    lockdep_is_held(&table->tb6_lock));
991 			struct fib6_info *new_leaf;
992 			if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
993 				new_leaf = fib6_find_prefix(net, table, fn);
994 				fib6_info_hold(new_leaf);
995 
996 				rcu_assign_pointer(fn->leaf, new_leaf);
997 				fib6_info_release(rt);
998 			}
999 			fn = rcu_dereference_protected(fn->parent,
1000 				    lockdep_is_held(&table->tb6_lock));
1001 		}
1002 	}
1003 }
1004 
1005 /*
1006  *	Insert routing information in a node.
1007  */
1008 
1009 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1010 			    struct nl_info *info,
1011 			    struct netlink_ext_ack *extack)
1012 {
1013 	struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1014 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1015 	struct fib6_info *iter = NULL;
1016 	struct fib6_info __rcu **ins;
1017 	struct fib6_info __rcu **fallback_ins = NULL;
1018 	int replace = (info->nlh &&
1019 		       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1020 	int add = (!info->nlh ||
1021 		   (info->nlh->nlmsg_flags & NLM_F_CREATE));
1022 	int found = 0;
1023 	bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1024 	u16 nlflags = NLM_F_EXCL;
1025 	int err;
1026 
1027 	if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1028 		nlflags |= NLM_F_APPEND;
1029 
1030 	ins = &fn->leaf;
1031 
1032 	for (iter = leaf; iter;
1033 	     iter = rcu_dereference_protected(iter->fib6_next,
1034 				lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1035 		/*
1036 		 *	Search for duplicates
1037 		 */
1038 
1039 		if (iter->fib6_metric == rt->fib6_metric) {
1040 			/*
1041 			 *	Same priority level
1042 			 */
1043 			if (info->nlh &&
1044 			    (info->nlh->nlmsg_flags & NLM_F_EXCL))
1045 				return -EEXIST;
1046 
1047 			nlflags &= ~NLM_F_EXCL;
1048 			if (replace) {
1049 				if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1050 					found++;
1051 					break;
1052 				}
1053 				if (rt_can_ecmp)
1054 					fallback_ins = fallback_ins ?: ins;
1055 				goto next_iter;
1056 			}
1057 
1058 			if (rt6_duplicate_nexthop(iter, rt)) {
1059 				if (rt->fib6_nsiblings)
1060 					rt->fib6_nsiblings = 0;
1061 				if (!(iter->fib6_flags & RTF_EXPIRES))
1062 					return -EEXIST;
1063 				if (!(rt->fib6_flags & RTF_EXPIRES))
1064 					fib6_clean_expires(iter);
1065 				else
1066 					fib6_set_expires(iter, rt->expires);
1067 
1068 				if (rt->fib6_pmtu)
1069 					fib6_metric_set(iter, RTAX_MTU,
1070 							rt->fib6_pmtu);
1071 				return -EEXIST;
1072 			}
1073 			/* If we have the same destination and the same metric,
1074 			 * but not the same gateway, then the route we try to
1075 			 * add is sibling to this route, increment our counter
1076 			 * of siblings, and later we will add our route to the
1077 			 * list.
1078 			 * Only static routes (which don't have flag
1079 			 * RTF_EXPIRES) are used for ECMPv6.
1080 			 *
1081 			 * To avoid long list, we only had siblings if the
1082 			 * route have a gateway.
1083 			 */
1084 			if (rt_can_ecmp &&
1085 			    rt6_qualify_for_ecmp(iter))
1086 				rt->fib6_nsiblings++;
1087 		}
1088 
1089 		if (iter->fib6_metric > rt->fib6_metric)
1090 			break;
1091 
1092 next_iter:
1093 		ins = &iter->fib6_next;
1094 	}
1095 
1096 	if (fallback_ins && !found) {
1097 		/* No ECMP-able route found, replace first non-ECMP one */
1098 		ins = fallback_ins;
1099 		iter = rcu_dereference_protected(*ins,
1100 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1101 		found++;
1102 	}
1103 
1104 	/* Reset round-robin state, if necessary */
1105 	if (ins == &fn->leaf)
1106 		fn->rr_ptr = NULL;
1107 
1108 	/* Link this route to others same route. */
1109 	if (rt->fib6_nsiblings) {
1110 		unsigned int fib6_nsiblings;
1111 		struct fib6_info *sibling, *temp_sibling;
1112 
1113 		/* Find the first route that have the same metric */
1114 		sibling = leaf;
1115 		while (sibling) {
1116 			if (sibling->fib6_metric == rt->fib6_metric &&
1117 			    rt6_qualify_for_ecmp(sibling)) {
1118 				list_add_tail(&rt->fib6_siblings,
1119 					      &sibling->fib6_siblings);
1120 				break;
1121 			}
1122 			sibling = rcu_dereference_protected(sibling->fib6_next,
1123 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1124 		}
1125 		/* For each sibling in the list, increment the counter of
1126 		 * siblings. BUG() if counters does not match, list of siblings
1127 		 * is broken!
1128 		 */
1129 		fib6_nsiblings = 0;
1130 		list_for_each_entry_safe(sibling, temp_sibling,
1131 					 &rt->fib6_siblings, fib6_siblings) {
1132 			sibling->fib6_nsiblings++;
1133 			BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1134 			fib6_nsiblings++;
1135 		}
1136 		BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1137 		rt6_multipath_rebalance(temp_sibling);
1138 	}
1139 
1140 	/*
1141 	 *	insert node
1142 	 */
1143 	if (!replace) {
1144 		if (!add)
1145 			pr_warn("NLM_F_CREATE should be set when creating new route\n");
1146 
1147 add:
1148 		nlflags |= NLM_F_CREATE;
1149 
1150 		if (!info->skip_notify_kernel) {
1151 			err = call_fib6_entry_notifiers(info->nl_net,
1152 							FIB_EVENT_ENTRY_ADD,
1153 							rt, extack);
1154 			if (err)
1155 				return err;
1156 		}
1157 
1158 		rcu_assign_pointer(rt->fib6_next, iter);
1159 		fib6_info_hold(rt);
1160 		rcu_assign_pointer(rt->fib6_node, fn);
1161 		rcu_assign_pointer(*ins, rt);
1162 		if (!info->skip_notify)
1163 			inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1164 		info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1165 
1166 		if (!(fn->fn_flags & RTN_RTINFO)) {
1167 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1168 			fn->fn_flags |= RTN_RTINFO;
1169 		}
1170 
1171 	} else {
1172 		int nsiblings;
1173 
1174 		if (!found) {
1175 			if (add)
1176 				goto add;
1177 			pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1178 			return -ENOENT;
1179 		}
1180 
1181 		if (!info->skip_notify_kernel) {
1182 			err = call_fib6_entry_notifiers(info->nl_net,
1183 							FIB_EVENT_ENTRY_REPLACE,
1184 							rt, extack);
1185 			if (err)
1186 				return err;
1187 		}
1188 
1189 		fib6_info_hold(rt);
1190 		rcu_assign_pointer(rt->fib6_node, fn);
1191 		rt->fib6_next = iter->fib6_next;
1192 		rcu_assign_pointer(*ins, rt);
1193 		if (!info->skip_notify)
1194 			inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1195 		if (!(fn->fn_flags & RTN_RTINFO)) {
1196 			info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1197 			fn->fn_flags |= RTN_RTINFO;
1198 		}
1199 		nsiblings = iter->fib6_nsiblings;
1200 		iter->fib6_node = NULL;
1201 		fib6_purge_rt(iter, fn, info->nl_net);
1202 		if (rcu_access_pointer(fn->rr_ptr) == iter)
1203 			fn->rr_ptr = NULL;
1204 		fib6_info_release(iter);
1205 
1206 		if (nsiblings) {
1207 			/* Replacing an ECMP route, remove all siblings */
1208 			ins = &rt->fib6_next;
1209 			iter = rcu_dereference_protected(*ins,
1210 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1211 			while (iter) {
1212 				if (iter->fib6_metric > rt->fib6_metric)
1213 					break;
1214 				if (rt6_qualify_for_ecmp(iter)) {
1215 					*ins = iter->fib6_next;
1216 					iter->fib6_node = NULL;
1217 					fib6_purge_rt(iter, fn, info->nl_net);
1218 					if (rcu_access_pointer(fn->rr_ptr) == iter)
1219 						fn->rr_ptr = NULL;
1220 					fib6_info_release(iter);
1221 					nsiblings--;
1222 					info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1223 				} else {
1224 					ins = &iter->fib6_next;
1225 				}
1226 				iter = rcu_dereference_protected(*ins,
1227 					lockdep_is_held(&rt->fib6_table->tb6_lock));
1228 			}
1229 			WARN_ON(nsiblings != 0);
1230 		}
1231 	}
1232 
1233 	return 0;
1234 }
1235 
1236 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1237 {
1238 	if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1239 	    (rt->fib6_flags & RTF_EXPIRES))
1240 		mod_timer(&net->ipv6.ip6_fib_timer,
1241 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1242 }
1243 
1244 void fib6_force_start_gc(struct net *net)
1245 {
1246 	if (!timer_pending(&net->ipv6.ip6_fib_timer))
1247 		mod_timer(&net->ipv6.ip6_fib_timer,
1248 			  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1249 }
1250 
1251 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1252 					   int sernum)
1253 {
1254 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1255 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1256 
1257 	/* paired with smp_rmb() in rt6_get_cookie_safe() */
1258 	smp_wmb();
1259 	while (fn) {
1260 		fn->fn_sernum = sernum;
1261 		fn = rcu_dereference_protected(fn->parent,
1262 				lockdep_is_held(&rt->fib6_table->tb6_lock));
1263 	}
1264 }
1265 
1266 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1267 {
1268 	__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1269 }
1270 
1271 /* allow ipv4 to update sernum via ipv6_stub */
1272 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1273 {
1274 	spin_lock_bh(&f6i->fib6_table->tb6_lock);
1275 	fib6_update_sernum_upto_root(net, f6i);
1276 	spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1277 }
1278 
1279 /*
1280  *	Add routing information to the routing tree.
1281  *	<destination addr>/<source addr>
1282  *	with source addr info in sub-trees
1283  *	Need to own table->tb6_lock
1284  */
1285 
1286 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1287 	     struct nl_info *info, struct netlink_ext_ack *extack)
1288 {
1289 	struct fib6_table *table = rt->fib6_table;
1290 	struct fib6_node *fn, *pn = NULL;
1291 	int err = -ENOMEM;
1292 	int allow_create = 1;
1293 	int replace_required = 0;
1294 	int sernum = fib6_new_sernum(info->nl_net);
1295 
1296 	if (info->nlh) {
1297 		if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1298 			allow_create = 0;
1299 		if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1300 			replace_required = 1;
1301 	}
1302 	if (!allow_create && !replace_required)
1303 		pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1304 
1305 	fn = fib6_add_1(info->nl_net, table, root,
1306 			&rt->fib6_dst.addr, rt->fib6_dst.plen,
1307 			offsetof(struct fib6_info, fib6_dst), allow_create,
1308 			replace_required, extack);
1309 	if (IS_ERR(fn)) {
1310 		err = PTR_ERR(fn);
1311 		fn = NULL;
1312 		goto out;
1313 	}
1314 
1315 	pn = fn;
1316 
1317 #ifdef CONFIG_IPV6_SUBTREES
1318 	if (rt->fib6_src.plen) {
1319 		struct fib6_node *sn;
1320 
1321 		if (!rcu_access_pointer(fn->subtree)) {
1322 			struct fib6_node *sfn;
1323 
1324 			/*
1325 			 * Create subtree.
1326 			 *
1327 			 *		fn[main tree]
1328 			 *		|
1329 			 *		sfn[subtree root]
1330 			 *		   \
1331 			 *		    sn[new leaf node]
1332 			 */
1333 
1334 			/* Create subtree root node */
1335 			sfn = node_alloc(info->nl_net);
1336 			if (!sfn)
1337 				goto failure;
1338 
1339 			fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1340 			rcu_assign_pointer(sfn->leaf,
1341 					   info->nl_net->ipv6.fib6_null_entry);
1342 			sfn->fn_flags = RTN_ROOT;
1343 
1344 			/* Now add the first leaf node to new subtree */
1345 
1346 			sn = fib6_add_1(info->nl_net, table, sfn,
1347 					&rt->fib6_src.addr, rt->fib6_src.plen,
1348 					offsetof(struct fib6_info, fib6_src),
1349 					allow_create, replace_required, extack);
1350 
1351 			if (IS_ERR(sn)) {
1352 				/* If it is failed, discard just allocated
1353 				   root, and then (in failure) stale node
1354 				   in main tree.
1355 				 */
1356 				node_free_immediate(info->nl_net, sfn);
1357 				err = PTR_ERR(sn);
1358 				goto failure;
1359 			}
1360 
1361 			/* Now link new subtree to main tree */
1362 			rcu_assign_pointer(sfn->parent, fn);
1363 			rcu_assign_pointer(fn->subtree, sfn);
1364 		} else {
1365 			sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1366 					&rt->fib6_src.addr, rt->fib6_src.plen,
1367 					offsetof(struct fib6_info, fib6_src),
1368 					allow_create, replace_required, extack);
1369 
1370 			if (IS_ERR(sn)) {
1371 				err = PTR_ERR(sn);
1372 				goto failure;
1373 			}
1374 		}
1375 
1376 		if (!rcu_access_pointer(fn->leaf)) {
1377 			if (fn->fn_flags & RTN_TL_ROOT) {
1378 				/* put back null_entry for root node */
1379 				rcu_assign_pointer(fn->leaf,
1380 					    info->nl_net->ipv6.fib6_null_entry);
1381 			} else {
1382 				fib6_info_hold(rt);
1383 				rcu_assign_pointer(fn->leaf, rt);
1384 			}
1385 		}
1386 		fn = sn;
1387 	}
1388 #endif
1389 
1390 	err = fib6_add_rt2node(fn, rt, info, extack);
1391 	if (!err) {
1392 		if (rt->nh)
1393 			list_add(&rt->nh_list, &rt->nh->f6i_list);
1394 		__fib6_update_sernum_upto_root(rt, sernum);
1395 		fib6_start_gc(info->nl_net, rt);
1396 	}
1397 
1398 out:
1399 	if (err) {
1400 #ifdef CONFIG_IPV6_SUBTREES
1401 		/*
1402 		 * If fib6_add_1 has cleared the old leaf pointer in the
1403 		 * super-tree leaf node we have to find a new one for it.
1404 		 */
1405 		if (pn != fn) {
1406 			struct fib6_info *pn_leaf =
1407 				rcu_dereference_protected(pn->leaf,
1408 				    lockdep_is_held(&table->tb6_lock));
1409 			if (pn_leaf == rt) {
1410 				pn_leaf = NULL;
1411 				RCU_INIT_POINTER(pn->leaf, NULL);
1412 				fib6_info_release(rt);
1413 			}
1414 			if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1415 				pn_leaf = fib6_find_prefix(info->nl_net, table,
1416 							   pn);
1417 #if RT6_DEBUG >= 2
1418 				if (!pn_leaf) {
1419 					WARN_ON(!pn_leaf);
1420 					pn_leaf =
1421 					    info->nl_net->ipv6.fib6_null_entry;
1422 				}
1423 #endif
1424 				fib6_info_hold(pn_leaf);
1425 				rcu_assign_pointer(pn->leaf, pn_leaf);
1426 			}
1427 		}
1428 #endif
1429 		goto failure;
1430 	}
1431 	return err;
1432 
1433 failure:
1434 	/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1435 	 * 1. fn is an intermediate node and we failed to add the new
1436 	 * route to it in both subtree creation failure and fib6_add_rt2node()
1437 	 * failure case.
1438 	 * 2. fn is the root node in the table and we fail to add the first
1439 	 * default route to it.
1440 	 */
1441 	if (fn &&
1442 	    (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1443 	     (fn->fn_flags & RTN_TL_ROOT &&
1444 	      !rcu_access_pointer(fn->leaf))))
1445 		fib6_repair_tree(info->nl_net, table, fn);
1446 	return err;
1447 }
1448 
1449 /*
1450  *	Routing tree lookup
1451  *
1452  */
1453 
1454 struct lookup_args {
1455 	int			offset;		/* key offset on fib6_info */
1456 	const struct in6_addr	*addr;		/* search key			*/
1457 };
1458 
1459 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1460 					    struct lookup_args *args)
1461 {
1462 	struct fib6_node *fn;
1463 	__be32 dir;
1464 
1465 	if (unlikely(args->offset == 0))
1466 		return NULL;
1467 
1468 	/*
1469 	 *	Descend on a tree
1470 	 */
1471 
1472 	fn = root;
1473 
1474 	for (;;) {
1475 		struct fib6_node *next;
1476 
1477 		dir = addr_bit_set(args->addr, fn->fn_bit);
1478 
1479 		next = dir ? rcu_dereference(fn->right) :
1480 			     rcu_dereference(fn->left);
1481 
1482 		if (next) {
1483 			fn = next;
1484 			continue;
1485 		}
1486 		break;
1487 	}
1488 
1489 	while (fn) {
1490 		struct fib6_node *subtree = FIB6_SUBTREE(fn);
1491 
1492 		if (subtree || fn->fn_flags & RTN_RTINFO) {
1493 			struct fib6_info *leaf = rcu_dereference(fn->leaf);
1494 			struct rt6key *key;
1495 
1496 			if (!leaf)
1497 				goto backtrack;
1498 
1499 			key = (struct rt6key *) ((u8 *)leaf + args->offset);
1500 
1501 			if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1502 #ifdef CONFIG_IPV6_SUBTREES
1503 				if (subtree) {
1504 					struct fib6_node *sfn;
1505 					sfn = fib6_node_lookup_1(subtree,
1506 								 args + 1);
1507 					if (!sfn)
1508 						goto backtrack;
1509 					fn = sfn;
1510 				}
1511 #endif
1512 				if (fn->fn_flags & RTN_RTINFO)
1513 					return fn;
1514 			}
1515 		}
1516 backtrack:
1517 		if (fn->fn_flags & RTN_ROOT)
1518 			break;
1519 
1520 		fn = rcu_dereference(fn->parent);
1521 	}
1522 
1523 	return NULL;
1524 }
1525 
1526 /* called with rcu_read_lock() held
1527  */
1528 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1529 				   const struct in6_addr *daddr,
1530 				   const struct in6_addr *saddr)
1531 {
1532 	struct fib6_node *fn;
1533 	struct lookup_args args[] = {
1534 		{
1535 			.offset = offsetof(struct fib6_info, fib6_dst),
1536 			.addr = daddr,
1537 		},
1538 #ifdef CONFIG_IPV6_SUBTREES
1539 		{
1540 			.offset = offsetof(struct fib6_info, fib6_src),
1541 			.addr = saddr,
1542 		},
1543 #endif
1544 		{
1545 			.offset = 0,	/* sentinel */
1546 		}
1547 	};
1548 
1549 	fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1550 	if (!fn || fn->fn_flags & RTN_TL_ROOT)
1551 		fn = root;
1552 
1553 	return fn;
1554 }
1555 
1556 /*
1557  *	Get node with specified destination prefix (and source prefix,
1558  *	if subtrees are used)
1559  *	exact_match == true means we try to find fn with exact match of
1560  *	the passed in prefix addr
1561  *	exact_match == false means we try to find fn with longest prefix
1562  *	match of the passed in prefix addr. This is useful for finding fn
1563  *	for cached route as it will be stored in the exception table under
1564  *	the node with longest prefix length.
1565  */
1566 
1567 
1568 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1569 				       const struct in6_addr *addr,
1570 				       int plen, int offset,
1571 				       bool exact_match)
1572 {
1573 	struct fib6_node *fn, *prev = NULL;
1574 
1575 	for (fn = root; fn ; ) {
1576 		struct fib6_info *leaf = rcu_dereference(fn->leaf);
1577 		struct rt6key *key;
1578 
1579 		/* This node is being deleted */
1580 		if (!leaf) {
1581 			if (plen <= fn->fn_bit)
1582 				goto out;
1583 			else
1584 				goto next;
1585 		}
1586 
1587 		key = (struct rt6key *)((u8 *)leaf + offset);
1588 
1589 		/*
1590 		 *	Prefix match
1591 		 */
1592 		if (plen < fn->fn_bit ||
1593 		    !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1594 			goto out;
1595 
1596 		if (plen == fn->fn_bit)
1597 			return fn;
1598 
1599 		if (fn->fn_flags & RTN_RTINFO)
1600 			prev = fn;
1601 
1602 next:
1603 		/*
1604 		 *	We have more bits to go
1605 		 */
1606 		if (addr_bit_set(addr, fn->fn_bit))
1607 			fn = rcu_dereference(fn->right);
1608 		else
1609 			fn = rcu_dereference(fn->left);
1610 	}
1611 out:
1612 	if (exact_match)
1613 		return NULL;
1614 	else
1615 		return prev;
1616 }
1617 
1618 struct fib6_node *fib6_locate(struct fib6_node *root,
1619 			      const struct in6_addr *daddr, int dst_len,
1620 			      const struct in6_addr *saddr, int src_len,
1621 			      bool exact_match)
1622 {
1623 	struct fib6_node *fn;
1624 
1625 	fn = fib6_locate_1(root, daddr, dst_len,
1626 			   offsetof(struct fib6_info, fib6_dst),
1627 			   exact_match);
1628 
1629 #ifdef CONFIG_IPV6_SUBTREES
1630 	if (src_len) {
1631 		WARN_ON(saddr == NULL);
1632 		if (fn) {
1633 			struct fib6_node *subtree = FIB6_SUBTREE(fn);
1634 
1635 			if (subtree) {
1636 				fn = fib6_locate_1(subtree, saddr, src_len,
1637 					   offsetof(struct fib6_info, fib6_src),
1638 					   exact_match);
1639 			}
1640 		}
1641 	}
1642 #endif
1643 
1644 	if (fn && fn->fn_flags & RTN_RTINFO)
1645 		return fn;
1646 
1647 	return NULL;
1648 }
1649 
1650 
1651 /*
1652  *	Deletion
1653  *
1654  */
1655 
1656 static struct fib6_info *fib6_find_prefix(struct net *net,
1657 					 struct fib6_table *table,
1658 					 struct fib6_node *fn)
1659 {
1660 	struct fib6_node *child_left, *child_right;
1661 
1662 	if (fn->fn_flags & RTN_ROOT)
1663 		return net->ipv6.fib6_null_entry;
1664 
1665 	while (fn) {
1666 		child_left = rcu_dereference_protected(fn->left,
1667 				    lockdep_is_held(&table->tb6_lock));
1668 		child_right = rcu_dereference_protected(fn->right,
1669 				    lockdep_is_held(&table->tb6_lock));
1670 		if (child_left)
1671 			return rcu_dereference_protected(child_left->leaf,
1672 					lockdep_is_held(&table->tb6_lock));
1673 		if (child_right)
1674 			return rcu_dereference_protected(child_right->leaf,
1675 					lockdep_is_held(&table->tb6_lock));
1676 
1677 		fn = FIB6_SUBTREE(fn);
1678 	}
1679 	return NULL;
1680 }
1681 
1682 /*
1683  *	Called to trim the tree of intermediate nodes when possible. "fn"
1684  *	is the node we want to try and remove.
1685  *	Need to own table->tb6_lock
1686  */
1687 
1688 static struct fib6_node *fib6_repair_tree(struct net *net,
1689 					  struct fib6_table *table,
1690 					  struct fib6_node *fn)
1691 {
1692 	int children;
1693 	int nstate;
1694 	struct fib6_node *child;
1695 	struct fib6_walker *w;
1696 	int iter = 0;
1697 
1698 	/* Set fn->leaf to null_entry for root node. */
1699 	if (fn->fn_flags & RTN_TL_ROOT) {
1700 		rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1701 		return fn;
1702 	}
1703 
1704 	for (;;) {
1705 		struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1706 					    lockdep_is_held(&table->tb6_lock));
1707 		struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1708 					    lockdep_is_held(&table->tb6_lock));
1709 		struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1710 					    lockdep_is_held(&table->tb6_lock));
1711 		struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1712 					    lockdep_is_held(&table->tb6_lock));
1713 		struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1714 					    lockdep_is_held(&table->tb6_lock));
1715 		struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1716 					    lockdep_is_held(&table->tb6_lock));
1717 		struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1718 					    lockdep_is_held(&table->tb6_lock));
1719 		struct fib6_info *new_fn_leaf;
1720 
1721 		RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1722 		iter++;
1723 
1724 		WARN_ON(fn->fn_flags & RTN_RTINFO);
1725 		WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1726 		WARN_ON(fn_leaf);
1727 
1728 		children = 0;
1729 		child = NULL;
1730 		if (fn_r)
1731 			child = fn_r, children |= 1;
1732 		if (fn_l)
1733 			child = fn_l, children |= 2;
1734 
1735 		if (children == 3 || FIB6_SUBTREE(fn)
1736 #ifdef CONFIG_IPV6_SUBTREES
1737 		    /* Subtree root (i.e. fn) may have one child */
1738 		    || (children && fn->fn_flags & RTN_ROOT)
1739 #endif
1740 		    ) {
1741 			new_fn_leaf = fib6_find_prefix(net, table, fn);
1742 #if RT6_DEBUG >= 2
1743 			if (!new_fn_leaf) {
1744 				WARN_ON(!new_fn_leaf);
1745 				new_fn_leaf = net->ipv6.fib6_null_entry;
1746 			}
1747 #endif
1748 			fib6_info_hold(new_fn_leaf);
1749 			rcu_assign_pointer(fn->leaf, new_fn_leaf);
1750 			return pn;
1751 		}
1752 
1753 #ifdef CONFIG_IPV6_SUBTREES
1754 		if (FIB6_SUBTREE(pn) == fn) {
1755 			WARN_ON(!(fn->fn_flags & RTN_ROOT));
1756 			RCU_INIT_POINTER(pn->subtree, NULL);
1757 			nstate = FWS_L;
1758 		} else {
1759 			WARN_ON(fn->fn_flags & RTN_ROOT);
1760 #endif
1761 			if (pn_r == fn)
1762 				rcu_assign_pointer(pn->right, child);
1763 			else if (pn_l == fn)
1764 				rcu_assign_pointer(pn->left, child);
1765 #if RT6_DEBUG >= 2
1766 			else
1767 				WARN_ON(1);
1768 #endif
1769 			if (child)
1770 				rcu_assign_pointer(child->parent, pn);
1771 			nstate = FWS_R;
1772 #ifdef CONFIG_IPV6_SUBTREES
1773 		}
1774 #endif
1775 
1776 		read_lock(&net->ipv6.fib6_walker_lock);
1777 		FOR_WALKERS(net, w) {
1778 			if (!child) {
1779 				if (w->node == fn) {
1780 					RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1781 					w->node = pn;
1782 					w->state = nstate;
1783 				}
1784 			} else {
1785 				if (w->node == fn) {
1786 					w->node = child;
1787 					if (children&2) {
1788 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1789 						w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1790 					} else {
1791 						RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1792 						w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1793 					}
1794 				}
1795 			}
1796 		}
1797 		read_unlock(&net->ipv6.fib6_walker_lock);
1798 
1799 		node_free(net, fn);
1800 		if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1801 			return pn;
1802 
1803 		RCU_INIT_POINTER(pn->leaf, NULL);
1804 		fib6_info_release(pn_leaf);
1805 		fn = pn;
1806 	}
1807 }
1808 
1809 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1810 			   struct fib6_info __rcu **rtp, struct nl_info *info)
1811 {
1812 	struct fib6_walker *w;
1813 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
1814 				    lockdep_is_held(&table->tb6_lock));
1815 	struct net *net = info->nl_net;
1816 
1817 	RT6_TRACE("fib6_del_route\n");
1818 
1819 	/* Unlink it */
1820 	*rtp = rt->fib6_next;
1821 	rt->fib6_node = NULL;
1822 	net->ipv6.rt6_stats->fib_rt_entries--;
1823 	net->ipv6.rt6_stats->fib_discarded_routes++;
1824 
1825 	/* Flush all cached dst in exception table */
1826 	rt6_flush_exceptions(rt);
1827 
1828 	/* Reset round-robin state, if necessary */
1829 	if (rcu_access_pointer(fn->rr_ptr) == rt)
1830 		fn->rr_ptr = NULL;
1831 
1832 	/* Remove this entry from other siblings */
1833 	if (rt->fib6_nsiblings) {
1834 		struct fib6_info *sibling, *next_sibling;
1835 
1836 		list_for_each_entry_safe(sibling, next_sibling,
1837 					 &rt->fib6_siblings, fib6_siblings)
1838 			sibling->fib6_nsiblings--;
1839 		rt->fib6_nsiblings = 0;
1840 		list_del_init(&rt->fib6_siblings);
1841 		rt6_multipath_rebalance(next_sibling);
1842 	}
1843 
1844 	/* Adjust walkers */
1845 	read_lock(&net->ipv6.fib6_walker_lock);
1846 	FOR_WALKERS(net, w) {
1847 		if (w->state == FWS_C && w->leaf == rt) {
1848 			RT6_TRACE("walker %p adjusted by delroute\n", w);
1849 			w->leaf = rcu_dereference_protected(rt->fib6_next,
1850 					    lockdep_is_held(&table->tb6_lock));
1851 			if (!w->leaf)
1852 				w->state = FWS_U;
1853 		}
1854 	}
1855 	read_unlock(&net->ipv6.fib6_walker_lock);
1856 
1857 	/* If it was last route, call fib6_repair_tree() to:
1858 	 * 1. For root node, put back null_entry as how the table was created.
1859 	 * 2. For other nodes, expunge its radix tree node.
1860 	 */
1861 	if (!rcu_access_pointer(fn->leaf)) {
1862 		if (!(fn->fn_flags & RTN_TL_ROOT)) {
1863 			fn->fn_flags &= ~RTN_RTINFO;
1864 			net->ipv6.rt6_stats->fib_route_nodes--;
1865 		}
1866 		fn = fib6_repair_tree(net, table, fn);
1867 	}
1868 
1869 	fib6_purge_rt(rt, fn, net);
1870 
1871 	if (!info->skip_notify_kernel)
1872 		call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1873 	if (!info->skip_notify)
1874 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1875 
1876 	fib6_info_release(rt);
1877 }
1878 
1879 /* Need to own table->tb6_lock */
1880 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1881 {
1882 	struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1883 				    lockdep_is_held(&rt->fib6_table->tb6_lock));
1884 	struct fib6_table *table = rt->fib6_table;
1885 	struct net *net = info->nl_net;
1886 	struct fib6_info __rcu **rtp;
1887 	struct fib6_info __rcu **rtp_next;
1888 
1889 	if (!fn || rt == net->ipv6.fib6_null_entry)
1890 		return -ENOENT;
1891 
1892 	WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1893 
1894 	/*
1895 	 *	Walk the leaf entries looking for ourself
1896 	 */
1897 
1898 	for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1899 		struct fib6_info *cur = rcu_dereference_protected(*rtp,
1900 					lockdep_is_held(&table->tb6_lock));
1901 		if (rt == cur) {
1902 			fib6_del_route(table, fn, rtp, info);
1903 			return 0;
1904 		}
1905 		rtp_next = &cur->fib6_next;
1906 	}
1907 	return -ENOENT;
1908 }
1909 
1910 /*
1911  *	Tree traversal function.
1912  *
1913  *	Certainly, it is not interrupt safe.
1914  *	However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1915  *	It means, that we can modify tree during walking
1916  *	and use this function for garbage collection, clone pruning,
1917  *	cleaning tree when a device goes down etc. etc.
1918  *
1919  *	It guarantees that every node will be traversed,
1920  *	and that it will be traversed only once.
1921  *
1922  *	Callback function w->func may return:
1923  *	0 -> continue walking.
1924  *	positive value -> walking is suspended (used by tree dumps,
1925  *	and probably by gc, if it will be split to several slices)
1926  *	negative value -> terminate walking.
1927  *
1928  *	The function itself returns:
1929  *	0   -> walk is complete.
1930  *	>0  -> walk is incomplete (i.e. suspended)
1931  *	<0  -> walk is terminated by an error.
1932  *
1933  *	This function is called with tb6_lock held.
1934  */
1935 
1936 static int fib6_walk_continue(struct fib6_walker *w)
1937 {
1938 	struct fib6_node *fn, *pn, *left, *right;
1939 
1940 	/* w->root should always be table->tb6_root */
1941 	WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1942 
1943 	for (;;) {
1944 		fn = w->node;
1945 		if (!fn)
1946 			return 0;
1947 
1948 		switch (w->state) {
1949 #ifdef CONFIG_IPV6_SUBTREES
1950 		case FWS_S:
1951 			if (FIB6_SUBTREE(fn)) {
1952 				w->node = FIB6_SUBTREE(fn);
1953 				continue;
1954 			}
1955 			w->state = FWS_L;
1956 #endif
1957 			/* fall through */
1958 		case FWS_L:
1959 			left = rcu_dereference_protected(fn->left, 1);
1960 			if (left) {
1961 				w->node = left;
1962 				w->state = FWS_INIT;
1963 				continue;
1964 			}
1965 			w->state = FWS_R;
1966 			/* fall through */
1967 		case FWS_R:
1968 			right = rcu_dereference_protected(fn->right, 1);
1969 			if (right) {
1970 				w->node = right;
1971 				w->state = FWS_INIT;
1972 				continue;
1973 			}
1974 			w->state = FWS_C;
1975 			w->leaf = rcu_dereference_protected(fn->leaf, 1);
1976 			/* fall through */
1977 		case FWS_C:
1978 			if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1979 				int err;
1980 
1981 				if (w->skip) {
1982 					w->skip--;
1983 					goto skip;
1984 				}
1985 
1986 				err = w->func(w);
1987 				if (err)
1988 					return err;
1989 
1990 				w->count++;
1991 				continue;
1992 			}
1993 skip:
1994 			w->state = FWS_U;
1995 			/* fall through */
1996 		case FWS_U:
1997 			if (fn == w->root)
1998 				return 0;
1999 			pn = rcu_dereference_protected(fn->parent, 1);
2000 			left = rcu_dereference_protected(pn->left, 1);
2001 			right = rcu_dereference_protected(pn->right, 1);
2002 			w->node = pn;
2003 #ifdef CONFIG_IPV6_SUBTREES
2004 			if (FIB6_SUBTREE(pn) == fn) {
2005 				WARN_ON(!(fn->fn_flags & RTN_ROOT));
2006 				w->state = FWS_L;
2007 				continue;
2008 			}
2009 #endif
2010 			if (left == fn) {
2011 				w->state = FWS_R;
2012 				continue;
2013 			}
2014 			if (right == fn) {
2015 				w->state = FWS_C;
2016 				w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2017 				continue;
2018 			}
2019 #if RT6_DEBUG >= 2
2020 			WARN_ON(1);
2021 #endif
2022 		}
2023 	}
2024 }
2025 
2026 static int fib6_walk(struct net *net, struct fib6_walker *w)
2027 {
2028 	int res;
2029 
2030 	w->state = FWS_INIT;
2031 	w->node = w->root;
2032 
2033 	fib6_walker_link(net, w);
2034 	res = fib6_walk_continue(w);
2035 	if (res <= 0)
2036 		fib6_walker_unlink(net, w);
2037 	return res;
2038 }
2039 
2040 static int fib6_clean_node(struct fib6_walker *w)
2041 {
2042 	int res;
2043 	struct fib6_info *rt;
2044 	struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2045 	struct nl_info info = {
2046 		.nl_net = c->net,
2047 		.skip_notify = c->skip_notify,
2048 	};
2049 
2050 	if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2051 	    w->node->fn_sernum != c->sernum)
2052 		w->node->fn_sernum = c->sernum;
2053 
2054 	if (!c->func) {
2055 		WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2056 		w->leaf = NULL;
2057 		return 0;
2058 	}
2059 
2060 	for_each_fib6_walker_rt(w) {
2061 		res = c->func(rt, c->arg);
2062 		if (res == -1) {
2063 			w->leaf = rt;
2064 			res = fib6_del(rt, &info);
2065 			if (res) {
2066 #if RT6_DEBUG >= 2
2067 				pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2068 					 __func__, rt,
2069 					 rcu_access_pointer(rt->fib6_node),
2070 					 res);
2071 #endif
2072 				continue;
2073 			}
2074 			return 0;
2075 		} else if (res == -2) {
2076 			if (WARN_ON(!rt->fib6_nsiblings))
2077 				continue;
2078 			rt = list_last_entry(&rt->fib6_siblings,
2079 					     struct fib6_info, fib6_siblings);
2080 			continue;
2081 		}
2082 		WARN_ON(res != 0);
2083 	}
2084 	w->leaf = rt;
2085 	return 0;
2086 }
2087 
2088 /*
2089  *	Convenient frontend to tree walker.
2090  *
2091  *	func is called on each route.
2092  *		It may return -2 -> skip multipath route.
2093  *			      -1 -> delete this route.
2094  *		              0  -> continue walking
2095  */
2096 
2097 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2098 			    int (*func)(struct fib6_info *, void *arg),
2099 			    int sernum, void *arg, bool skip_notify)
2100 {
2101 	struct fib6_cleaner c;
2102 
2103 	c.w.root = root;
2104 	c.w.func = fib6_clean_node;
2105 	c.w.count = 0;
2106 	c.w.skip = 0;
2107 	c.w.skip_in_node = 0;
2108 	c.func = func;
2109 	c.sernum = sernum;
2110 	c.arg = arg;
2111 	c.net = net;
2112 	c.skip_notify = skip_notify;
2113 
2114 	fib6_walk(net, &c.w);
2115 }
2116 
2117 static void __fib6_clean_all(struct net *net,
2118 			     int (*func)(struct fib6_info *, void *),
2119 			     int sernum, void *arg, bool skip_notify)
2120 {
2121 	struct fib6_table *table;
2122 	struct hlist_head *head;
2123 	unsigned int h;
2124 
2125 	rcu_read_lock();
2126 	for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2127 		head = &net->ipv6.fib_table_hash[h];
2128 		hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2129 			spin_lock_bh(&table->tb6_lock);
2130 			fib6_clean_tree(net, &table->tb6_root,
2131 					func, sernum, arg, skip_notify);
2132 			spin_unlock_bh(&table->tb6_lock);
2133 		}
2134 	}
2135 	rcu_read_unlock();
2136 }
2137 
2138 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2139 		    void *arg)
2140 {
2141 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2142 }
2143 
2144 void fib6_clean_all_skip_notify(struct net *net,
2145 				int (*func)(struct fib6_info *, void *),
2146 				void *arg)
2147 {
2148 	__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2149 }
2150 
2151 static void fib6_flush_trees(struct net *net)
2152 {
2153 	int new_sernum = fib6_new_sernum(net);
2154 
2155 	__fib6_clean_all(net, NULL, new_sernum, NULL, false);
2156 }
2157 
2158 /*
2159  *	Garbage collection
2160  */
2161 
2162 static int fib6_age(struct fib6_info *rt, void *arg)
2163 {
2164 	struct fib6_gc_args *gc_args = arg;
2165 	unsigned long now = jiffies;
2166 
2167 	/*
2168 	 *	check addrconf expiration here.
2169 	 *	Routes are expired even if they are in use.
2170 	 */
2171 
2172 	if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2173 		if (time_after(now, rt->expires)) {
2174 			RT6_TRACE("expiring %p\n", rt);
2175 			return -1;
2176 		}
2177 		gc_args->more++;
2178 	}
2179 
2180 	/*	Also age clones in the exception table.
2181 	 *	Note, that clones are aged out
2182 	 *	only if they are not in use now.
2183 	 */
2184 	rt6_age_exceptions(rt, gc_args, now);
2185 
2186 	return 0;
2187 }
2188 
2189 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2190 {
2191 	struct fib6_gc_args gc_args;
2192 	unsigned long now;
2193 
2194 	if (force) {
2195 		spin_lock_bh(&net->ipv6.fib6_gc_lock);
2196 	} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2197 		mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2198 		return;
2199 	}
2200 	gc_args.timeout = expires ? (int)expires :
2201 			  net->ipv6.sysctl.ip6_rt_gc_interval;
2202 	gc_args.more = 0;
2203 
2204 	fib6_clean_all(net, fib6_age, &gc_args);
2205 	now = jiffies;
2206 	net->ipv6.ip6_rt_last_gc = now;
2207 
2208 	if (gc_args.more)
2209 		mod_timer(&net->ipv6.ip6_fib_timer,
2210 			  round_jiffies(now
2211 					+ net->ipv6.sysctl.ip6_rt_gc_interval));
2212 	else
2213 		del_timer(&net->ipv6.ip6_fib_timer);
2214 	spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2215 }
2216 
2217 static void fib6_gc_timer_cb(struct timer_list *t)
2218 {
2219 	struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2220 
2221 	fib6_run_gc(0, arg, true);
2222 }
2223 
2224 static int __net_init fib6_net_init(struct net *net)
2225 {
2226 	size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2227 	int err;
2228 
2229 	err = fib6_notifier_init(net);
2230 	if (err)
2231 		return err;
2232 
2233 	spin_lock_init(&net->ipv6.fib6_gc_lock);
2234 	rwlock_init(&net->ipv6.fib6_walker_lock);
2235 	INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2236 	timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2237 
2238 	net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2239 	if (!net->ipv6.rt6_stats)
2240 		goto out_timer;
2241 
2242 	/* Avoid false sharing : Use at least a full cache line */
2243 	size = max_t(size_t, size, L1_CACHE_BYTES);
2244 
2245 	net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2246 	if (!net->ipv6.fib_table_hash)
2247 		goto out_rt6_stats;
2248 
2249 	net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2250 					  GFP_KERNEL);
2251 	if (!net->ipv6.fib6_main_tbl)
2252 		goto out_fib_table_hash;
2253 
2254 	net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2255 	rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2256 			   net->ipv6.fib6_null_entry);
2257 	net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2258 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2259 	inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2260 
2261 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2262 	net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2263 					   GFP_KERNEL);
2264 	if (!net->ipv6.fib6_local_tbl)
2265 		goto out_fib6_main_tbl;
2266 	net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2267 	rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2268 			   net->ipv6.fib6_null_entry);
2269 	net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2270 		RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2271 	inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2272 #endif
2273 	fib6_tables_init(net);
2274 
2275 	return 0;
2276 
2277 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2278 out_fib6_main_tbl:
2279 	kfree(net->ipv6.fib6_main_tbl);
2280 #endif
2281 out_fib_table_hash:
2282 	kfree(net->ipv6.fib_table_hash);
2283 out_rt6_stats:
2284 	kfree(net->ipv6.rt6_stats);
2285 out_timer:
2286 	fib6_notifier_exit(net);
2287 	return -ENOMEM;
2288 }
2289 
2290 static void fib6_net_exit(struct net *net)
2291 {
2292 	unsigned int i;
2293 
2294 	del_timer_sync(&net->ipv6.ip6_fib_timer);
2295 
2296 	for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2297 		struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2298 		struct hlist_node *tmp;
2299 		struct fib6_table *tb;
2300 
2301 		hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2302 			hlist_del(&tb->tb6_hlist);
2303 			fib6_free_table(tb);
2304 		}
2305 	}
2306 
2307 	kfree(net->ipv6.fib_table_hash);
2308 	kfree(net->ipv6.rt6_stats);
2309 	fib6_notifier_exit(net);
2310 }
2311 
2312 static struct pernet_operations fib6_net_ops = {
2313 	.init = fib6_net_init,
2314 	.exit = fib6_net_exit,
2315 };
2316 
2317 int __init fib6_init(void)
2318 {
2319 	int ret = -ENOMEM;
2320 
2321 	fib6_node_kmem = kmem_cache_create("fib6_nodes",
2322 					   sizeof(struct fib6_node),
2323 					   0, SLAB_HWCACHE_ALIGN,
2324 					   NULL);
2325 	if (!fib6_node_kmem)
2326 		goto out;
2327 
2328 	ret = register_pernet_subsys(&fib6_net_ops);
2329 	if (ret)
2330 		goto out_kmem_cache_create;
2331 
2332 	ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2333 				   inet6_dump_fib, 0);
2334 	if (ret)
2335 		goto out_unregister_subsys;
2336 
2337 	__fib6_flush_trees = fib6_flush_trees;
2338 out:
2339 	return ret;
2340 
2341 out_unregister_subsys:
2342 	unregister_pernet_subsys(&fib6_net_ops);
2343 out_kmem_cache_create:
2344 	kmem_cache_destroy(fib6_node_kmem);
2345 	goto out;
2346 }
2347 
2348 void fib6_gc_cleanup(void)
2349 {
2350 	unregister_pernet_subsys(&fib6_net_ops);
2351 	kmem_cache_destroy(fib6_node_kmem);
2352 }
2353 
2354 #ifdef CONFIG_PROC_FS
2355 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2356 {
2357 	struct fib6_info *rt = v;
2358 	struct ipv6_route_iter *iter = seq->private;
2359 	struct fib6_nh *fib6_nh = rt->fib6_nh;
2360 	unsigned int flags = rt->fib6_flags;
2361 	const struct net_device *dev;
2362 
2363 	if (rt->nh)
2364 		fib6_nh = nexthop_fib6_nh(rt->nh);
2365 
2366 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2367 
2368 #ifdef CONFIG_IPV6_SUBTREES
2369 	seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2370 #else
2371 	seq_puts(seq, "00000000000000000000000000000000 00 ");
2372 #endif
2373 	if (fib6_nh->fib_nh_gw_family) {
2374 		flags |= RTF_GATEWAY;
2375 		seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2376 	} else {
2377 		seq_puts(seq, "00000000000000000000000000000000");
2378 	}
2379 
2380 	dev = fib6_nh->fib_nh_dev;
2381 	seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2382 		   rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2383 		   flags, dev ? dev->name : "");
2384 	iter->w.leaf = NULL;
2385 	return 0;
2386 }
2387 
2388 static int ipv6_route_yield(struct fib6_walker *w)
2389 {
2390 	struct ipv6_route_iter *iter = w->args;
2391 
2392 	if (!iter->skip)
2393 		return 1;
2394 
2395 	do {
2396 		iter->w.leaf = rcu_dereference_protected(
2397 				iter->w.leaf->fib6_next,
2398 				lockdep_is_held(&iter->tbl->tb6_lock));
2399 		iter->skip--;
2400 		if (!iter->skip && iter->w.leaf)
2401 			return 1;
2402 	} while (iter->w.leaf);
2403 
2404 	return 0;
2405 }
2406 
2407 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2408 				      struct net *net)
2409 {
2410 	memset(&iter->w, 0, sizeof(iter->w));
2411 	iter->w.func = ipv6_route_yield;
2412 	iter->w.root = &iter->tbl->tb6_root;
2413 	iter->w.state = FWS_INIT;
2414 	iter->w.node = iter->w.root;
2415 	iter->w.args = iter;
2416 	iter->sernum = iter->w.root->fn_sernum;
2417 	INIT_LIST_HEAD(&iter->w.lh);
2418 	fib6_walker_link(net, &iter->w);
2419 }
2420 
2421 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2422 						    struct net *net)
2423 {
2424 	unsigned int h;
2425 	struct hlist_node *node;
2426 
2427 	if (tbl) {
2428 		h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2429 		node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2430 	} else {
2431 		h = 0;
2432 		node = NULL;
2433 	}
2434 
2435 	while (!node && h < FIB6_TABLE_HASHSZ) {
2436 		node = rcu_dereference_bh(
2437 			hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2438 	}
2439 	return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2440 }
2441 
2442 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2443 {
2444 	if (iter->sernum != iter->w.root->fn_sernum) {
2445 		iter->sernum = iter->w.root->fn_sernum;
2446 		iter->w.state = FWS_INIT;
2447 		iter->w.node = iter->w.root;
2448 		WARN_ON(iter->w.skip);
2449 		iter->w.skip = iter->w.count;
2450 	}
2451 }
2452 
2453 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2454 {
2455 	int r;
2456 	struct fib6_info *n;
2457 	struct net *net = seq_file_net(seq);
2458 	struct ipv6_route_iter *iter = seq->private;
2459 
2460 	if (!v)
2461 		goto iter_table;
2462 
2463 	n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2464 	if (n) {
2465 		++*pos;
2466 		return n;
2467 	}
2468 
2469 iter_table:
2470 	ipv6_route_check_sernum(iter);
2471 	spin_lock_bh(&iter->tbl->tb6_lock);
2472 	r = fib6_walk_continue(&iter->w);
2473 	spin_unlock_bh(&iter->tbl->tb6_lock);
2474 	if (r > 0) {
2475 		if (v)
2476 			++*pos;
2477 		return iter->w.leaf;
2478 	} else if (r < 0) {
2479 		fib6_walker_unlink(net, &iter->w);
2480 		return NULL;
2481 	}
2482 	fib6_walker_unlink(net, &iter->w);
2483 
2484 	iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2485 	if (!iter->tbl)
2486 		return NULL;
2487 
2488 	ipv6_route_seq_setup_walk(iter, net);
2489 	goto iter_table;
2490 }
2491 
2492 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2493 	__acquires(RCU_BH)
2494 {
2495 	struct net *net = seq_file_net(seq);
2496 	struct ipv6_route_iter *iter = seq->private;
2497 
2498 	rcu_read_lock_bh();
2499 	iter->tbl = ipv6_route_seq_next_table(NULL, net);
2500 	iter->skip = *pos;
2501 
2502 	if (iter->tbl) {
2503 		ipv6_route_seq_setup_walk(iter, net);
2504 		return ipv6_route_seq_next(seq, NULL, pos);
2505 	} else {
2506 		return NULL;
2507 	}
2508 }
2509 
2510 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2511 {
2512 	struct fib6_walker *w = &iter->w;
2513 	return w->node && !(w->state == FWS_U && w->node == w->root);
2514 }
2515 
2516 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2517 	__releases(RCU_BH)
2518 {
2519 	struct net *net = seq_file_net(seq);
2520 	struct ipv6_route_iter *iter = seq->private;
2521 
2522 	if (ipv6_route_iter_active(iter))
2523 		fib6_walker_unlink(net, &iter->w);
2524 
2525 	rcu_read_unlock_bh();
2526 }
2527 
2528 const struct seq_operations ipv6_route_seq_ops = {
2529 	.start	= ipv6_route_seq_start,
2530 	.next	= ipv6_route_seq_next,
2531 	.stop	= ipv6_route_seq_stop,
2532 	.show	= ipv6_route_seq_show
2533 };
2534 #endif /* CONFIG_PROC_FS */
2535